250128 - Gaussian Splating 3 por Mario Rocha

00:00:00

the minutes before the junk dealer, excellent

00:00:05

Well, we live in a time for this, don't we? Yes.

00:00:08

Well yes, I'm ready with some of this Gau shen stuff here.

00:00:14

Something that

00:00:16

are interested, that

00:00:17

I like it here as first, like reviewing, so very briefly, right? Well, look, I think that

00:00:21

they do it like that without

00:00:22

You're the main one, aren't you? The fight with which

00:00:24

already found me, see me facing her now, because

00:00:26

Of course, I notice the difference between the theory, having seen it, having seen your examples, your work, but seeing myself already faced in the field with the issue of scanning, because

00:00:39

For example, with Oscar, we sometimes had different visions of how to scan.

00:00:45

Tell me! I can't hear you, you're too tired.

00:00:48

No, you hear me.

00:00:49

No, I know, I hear it.

00:00:51

Yes, also

00:00:52

Tell him that

00:00:53

closing and refreshment the same

00:00:55

It's my thing

00:00:57

That's why he didn't listen to me before.

00:01:06

I say that

00:01:08

for example Oscar, for example, the drone that

00:01:13

It does not have to point the camera upwards, which is

00:01:16

I see it as necessary when there are ceilings or overhangs, or details below, as was the specific case in my work, and above all to have that scheme

00:01:30

scanning and what

00:01:32

Oscar told me, in professional jobs you need to recognize the site to know how to proceed when scanning it, right? So, specifically, my first job, well, it's a horror of buildings, a

00:01:47

beautiful, but a horror when it comes to resting it, because

00:01:50

You can't make a complete orbit, you can't put a programmed drone link on it, because

00:01:58

has a house next door in which

00:02:00

you can crash the drone, because

00:02:03

the same home retort can be at the other end and suddenly on returning it collides with the house, because

00:02:12

It is quite higher than

00:02:13

the building itself

00:02:14

So, yes,

00:02:16

Well, we marked the scan a little, as I sent you the capture, and you see that

00:02:22

There are the camera positions, I did it from above to capture the entire top, I did it at 45 to 30 degrees downwards to have that diagonal and the floors inside the gallery, and then

00:02:39

at minus 30 and up to catch the ceilings of the interiors of the galleries only inside the galleries, which

00:02:47

I think that

00:02:48

I should have mapped the facade again from below too, but well, that would make the project more complicated.

00:02:55

So, I would mainly like to talk about the scanning, which

00:02:59

I don't think so

00:03:00

We haven't talked much and that's where I've also seen a problem, and then I think that

00:03:04

also something that

00:03:05

Of great interest is the workflow, from scanning to processing and publishing.

00:03:12

Mario? Yes, hello, Leon, how are you all? Hello, no, I wanted to too.

00:03:18

add a little bit like the specific thing that

00:03:21

could do, such as the needs or concerns that

00:03:25

We have with the workflow part, last week I was helping to take, do a scan of furniture, for testing, and of course, one of the main problems that

00:03:34

They don't have here when trying to do tests are the times, because

00:03:38

For example, you might want to use one type of lighting or another, or record a video that

00:03:44

You can turn it two times, another few times, change the object or change the angles, and each of those tests has to

00:03:49

go through a whole process that

00:03:51

It is quite slow to have final results and compare

00:03:53

So, I think that

00:03:55

It would be super useful to share everyone's experiences to try to find good ideas and good solutions when it comes to

00:04:02

the scans, for example, be as good as possible

00:04:05

have the right lighting, angles and everything, I think

00:04:08

that has a lot to do with

00:04:09

see with making mistakes and with experience as in the practice of having performed the scans

00:04:14

And the other part that

00:04:16

I loved you too

00:04:17

as if to ask out of curiosity, how is this matter going, is that

00:04:21

When it comes to the optimization part, I have been working now in Blender trying mainly to take scans and realize what the results are, for example, in Kiri it gives me the possibility that

00:04:33

It generates a met or that

00:04:34

It generates a splat, and then I have seen different results, varied depending on the type of geometry of the objects, and I have made challenges for legislation and texture recovery, and there are times that

00:04:47

It's super simple, sometimes that

00:04:48

The results are quite, well, again that

00:04:51

more complicated, I think

00:04:52

It is good to start understanding when one proposes a job that

00:04:56

you will need that post-scan optimization so much, and also in the export part, because

00:05:04

at the time of exporting and I realize that

00:05:05

There is a lot of difference in file sizes, so when we work with small things, but repeatedly, for example, that

00:05:12

I know it is furniture, it matters, or when working with large objects, like that

00:05:16

I know a view and a city that

00:05:18

have a lot of polygons, I also care about them, so I've been doing some preas, in fact, one of the things that

00:05:25

I realized that

00:05:26

In SuperSplat the export gives you some compressed formats, which

00:05:30

because you don't take notice, and I also realize that

00:05:33

The voucher format in Splat, I downloaded these papers, tried to read them and kept informed about them, it is in full development, and there are so many changes, so emerging, that

00:05:43

There are a lot of things that

00:05:44

They are mainly trying out these motion captures, trying to capture the splats moving, which

00:05:49

not probably what

00:05:50

let's talk now in a couple of months it will change, because

00:05:54

It is not yet standardized, and that is a problem or a challenge for not doing it, so I would also like to see what this assessment is about it.

00:06:00

Yes, thanks

00:06:03

Something that

00:06:04

you mention it is, and it is important to touch it, right? That is so that

00:06:09

We're going to use it, aren't we? If you're talking about optimization and Blender, possibly the fact that

00:06:15

Let's work with gavosh and splat, then it's not the path, but the photography itself, starting there as on that part, subject to what

00:06:22

Kiri does, that

00:06:23

Kiri offers you both, he offers you the gavosh and he offers you this mesh, but he will have to

00:06:29

see a lot with the lens, right now for example, and we have been searching a lot with gavosh and splat mainly at three views, the main one, and secondly, using three max studies, which

00:06:43

is the software that

00:06:43

we use, that

00:06:44

It's similar to Blender, so you can integrate with real texts, with general buildings, right? But yes, I do think that

00:06:52

I'm beginning to understand where I'm coming from, that

00:06:54

I have material prepared right now, it's practical, and we're going to add to it, right? So if someone does this, they have to

00:07:00

put some ideas, I want to see if we're going to have a little bit of your stuff, right? And any questions, or something specific about Gavosh and

00:07:10

Yes, that's also optimization, right? Because

00:07:13

For example, I have had to

00:07:14

put, I did a test the other day at two thousand splats in the supermarket, in the hotshot, and see, when it comes to uploading it to three views, if only with what

00:07:25

you increased, to raise it to super splat and lower it again, or see some process so that

00:07:31

How will it go online, first have the highest quality possible, but that

00:07:35

then at the time of weighing we can upload it online

00:07:39

Okay, I'm bringing some of that, I see and also what

00:07:43

It happened to you in relation to the components, right? So, I say, it's not a solution yet, but it's a clue, very well, then I don't know, someone else, or I can share the screen, what do you think? Then you

00:08:01

He jumped at asking her, didn't he? Okay, okay.

00:08:04

Yes, I'm going to try to measure my times, not at the best, this part, more theoretical, about 10 minutes, if we go to something practical so that

00:08:13

be more substantial, as per Pietro's request, I will try to speak more slowly, okay, very good, excellent

00:08:25

So, I share the screen and I share everything better, well, I don't know if you can tell me if you see everything correctly, it looks perfect, you can see it by moving the screen

00:08:42

Excellent, I'm going to dedicate 10 minutes to catch up, if you don't mind, in relation to the software that

00:08:49

There, everything is focused on Gaussian Splatting and well, and after that, we go like a case study where we could open some data set to see the scan, more or less like in my experience,

00:09:05

how it has turned out for me, and on the other hand, the optimization part, that

00:09:13

Well, we had technically lost Super Splat to use only PostShot, but this one, I think that

00:09:21

We can now use Super Splat again and three views are lined up for the moment, as Leon said, everything is still in beta, this whole process is still in beta, so it's like what

00:09:31

I was talking at the beginning about GPT chat and this new phenomenon, no, Chinese, that

00:09:36

There, everything is changing and it can change completely from one day to the next, so well, I'll tell you a little bit about what

00:09:45

I bring it here, remember that

00:09:47

You have access, I don't know if you can confirm, you have access to this document, which

00:09:54

Well, I always keep feeding, where here they can move some things that

00:10:00

are already obsolete, others that

00:10:02

Well, they are in the spotlight, like the new existing formats, and the roadmap is always starred, which

00:10:10

It's more or less how it is developing and also where this technology is extending to. Well, you can explore it here, so as not to go into much detail in that part, the reconstruction methods, nothing more.

00:10:25

As a recap, there are two main ones, one, that

00:10:31

It's, I don't have time there, Leon mentioned to me just now, how can we verify this quickly, no, before leaving the place, I don't know, imagine that

00:10:40

is on location, taking many photographs, they will be there for two days, well, obviously working with the post-shot methodology would undoubtedly take a lot of time, especially if we are out of our office, and not

00:10:54

We have a computer, a powerful computer, well, what

00:10:58

I have implemented is to use some of these tools, well, not this one, what

00:11:05

It would be the luma, which

00:11:06

Unfortunately it is very saturated, so as you have noticed, those who use it, well it can last up to three days and give you a result, you upload your catch and three days later you are receiving when before it happened in

00:11:22

minutes this, is already as if it were over, it is still a good option within the solutions that

00:11:29

We have the cloud to be uploading information even from our mobile, the drone connected to the mobile and from the Aluma mobile, that is a possibility to have, it was a possibility to have a quick, preliminary result before

00:11:43

If I know, let's imagine that

00:11:48

They scan the façade with some overhead shots and some orbits to be able to foresee any problems, so Luma was like a possibility, another possibility is Policam and Kiri, both are paid, I think.

00:12:04

Policam has given me slightly faster results than

00:12:10

Kiri, but they are not like that anymore, a scan can take a couple of hours to resolve, unlike Luma, which

00:12:25

They give it to you in three days, so maybe in those couple of hours you could implement some solution to upload the information, continue taking pictures and have a preliminary before leaving the place thinking about

00:12:40

what

00:12:40

They have to travel to be able to take photos of a building, something that

00:12:45

I have not tried it and I also want to try Kiri on the web, which

00:12:50

It is the methodology under these tools that

00:12:54

process in the cloud, which

00:12:56

It is practically allowing us to upload from your website that

00:13:01

It can be from our mobile phone, since it is a lot of information to be able to obtain this.

00:13:07

I mentioned Leon, right now we have this double in Kiri that

00:13:10

we have what

00:13:11

It is the Gaussian and the Maya, both for the price of one, that is, we upload the image dataset and in the end we obtain a Maya and a Gaussian but via the web, which

00:13:23

It is the way in which

00:13:25

we could bring us information that

00:13:28

It does not leave our mobile phone, but

00:13:29

They can be drone photos, photos from several cameras, 360 cameras among others, so this is the reconstruction methodology that

00:13:40

We already know them all, nothing more, well it was like taking them up again a little bit, I would recommend them since

00:13:47

The Luma is a little saturated, maybe make an effort or together try Kiri and the Polycam to see which one is currently doing more efficiently in these scans, express, let's call it, a tip

00:14:01

Mario, before that

00:14:02

you already give up, tell you that

00:14:04

The Luma has improved, I think.

00:14:06

They did some maintenance and it has improved because

00:14:09

a test that

00:14:11

I did it, he did it quickly, he also did it in one or two hours and I was surprised, because

00:14:16

had been getting stuck for a while

00:14:18

I did it three days, but they must have done some maintenance or something, but it doesn't take away, as I have said on several occasions, that

00:14:25

Being free for some toll, we will see what happens present or future and one of them is that

00:14:32

People are getting into the free one and it could obviously become saturated again and if it's your only tool you're going to depend on it, then it could leave you stranded and then I don't know if Polycam

00:14:44

and Kiri allows you to video like Luma allows you to shoot raw video through Polycam only images nothing more and here it seems to me that

00:14:57

in Kiri only images too, in case it is different from your little that

00:15:05

luma the truth that

00:15:06

is very comfortable because

00:15:07

allows you both

00:15:09

Exact

00:15:10

If luma is much more complete and more streamed, it is good information that

00:15:15

you share with us, that

00:15:17

they are already improving

00:15:18

We tried to search to see if there was any documentation, since

00:15:21

We have spot-shot of how luma processes, which

00:15:25

has prey when throwing material at him, which

00:15:29

prese has of the splats, of the KST, of these stories, like the spot-shot we can see the guts, I have not found anywhere the documentation of luma, that

00:15:41

Tell me how he does the prese, because

00:15:43

Since it doesn't let you configure, at least know what applies.

00:15:47

I haven't found it

00:15:48

Yes, correct, I say, here in those three methodologies you save the CPU, the processing, but you lose control over what

00:15:59

is happening with the scanning

00:16:01

Understand that

00:16:02

they

00:16:03

Mario, can you upload a video? Up to three minutes for the pro account and one minute for the free account.

00:16:10

And if anyone wants to do some testing, I have had my account turned off for many months and I wanted to,

00:16:16

so

00:16:17

They just contact me and try it out.

00:16:20

Thank you very much, I also have the account, it is paid for, but I have not tried it, I will check, maybe here, when I change to Gaussion, it will give me the option, but here I only made jpjgif, I don't know

00:16:35

if something is hidden there

00:16:37

It's over if you look at Kiride and the photo and video formats are allowed.

00:16:43

Ah, correct.

00:16:44

Yes, yes, thank you

00:16:45

Yeah, so we corrected that, then Kiri, if it would also allow us to upload video, which is very similar to luma, right? Nothing more than,

00:16:56

Well, the video limit is one minute and three minutes in pro, which is very little, Luma allows us to upload 15 minutes of video, so yes, even though

00:17:08

It's a pro version, it has its limits, right? And photos also seemed to me like 300 photos, it seemed like a limit to me because they were very few, very few, right? Kiri is very focused on the object just in case, by

00:17:22

now

00:17:23

Sure, sure

00:17:25

If not, here, for example, Policam, just to have the data, also allows us to upload 15 minutes or 16 video streams.

00:17:33

And a thousand photos

00:17:33

Yes, that,

00:17:34

dog

00:17:35

And a thousand photos in the pro version, obviously, right? So it's much better.

00:17:40

And Luma, well, here, since

00:17:43

Yes, we all know it, Luma is something similar.

00:17:46

It is so powerful, because

00:17:47

the thing is

00:17:47

I don't understand that

00:17:48

is free, because,

00:17:50

I think that

00:17:50

is going to change, because

00:17:51

Look, I looked again at what

00:17:52

You taught us there about the help and it allows you to throw three equirectangular ones.

00:17:56

It's the, I don't know, I think I'll go out with this solid camera and scan

00:18:02

Yes, even in a row, look, it allows me to upload photos in a row, in HDR, well, yes, it seems much, much more powerful to me, but well, here the maximum video is 5 leagues

00:18:16

And in the case of Luma, suddenly from Policam, they're sending us 16 video arrivals here, right? Of course.

00:18:24

Which, well, we could practically solve with a video compressor or something like that, right? To be able to upload more content.

00:18:33

Anyway, I think that

00:18:34

You will surely get there later, but the process is that

00:18:38

You discovered, using potshot, to select the frames, both exporting them and the best ones

00:18:44

I think that

00:18:45

Here we start to forget whether the video weighs 15 you arrive or not, because

00:18:49

With that intermediate step you speed up the process a lot.

00:18:54

Of course, well, yes, yes, yes, right now we'll touch on it very quickly, right? To get more practical, right? So, in summary, we have these tools that

00:19:04

Many of us already know that

00:19:05

there is a new tool that

00:19:13

is emerging, both to explore it, and

00:19:16

more pel, it's still in, we're on the waiting list

00:19:21

It's interesting, because

00:19:22

Here you can create 3D models from an image or from a text like everything you

00:19:29

It's happening right now with artificial interest, but let's think that

00:19:33

3rd view is promising in the first quarter, the first quarter of the year, to have this own 3rd view tool, where they will allow you to also mix models, so, let's imagine that

00:19:46

We have a scan and you guys, just for a creative aspect, would like to maybe put in a, I don't know, you guys, like you call a witch, a wizard, then you basically download this wizard and merge him with

00:20:02

your scan, right? I'm thinking about what

00:20:05

It came to me here first, but let's think of a tree house or something else, merging it with creating, then this starts to open up in that sense, in the mix of models and content as is already happening.

00:20:20

with what

00:20:22

It's super splat, which

00:20:23

We're going to play it now, right? Then another of the directions that

00:20:28

It can be taken for this type of reconstructions with Gaussian, since it is the part of the reconstruction in local mode, right? On their computers, using video processing mainly, let's say that

00:20:42

Technology emerges, right with Nerf Studio, here you can explore what

00:20:47

It is about, it is still valid and Nerf is a technology prior to Gaussian splatting, however, it has been generated as a hybrid, it is very interesting, but well, here you require code and a paper that

00:21:02

Well, you can install it on your computer, you'll notice a lot, but the process is still slow, right? However, it's much more fluid at a professional level, this could even be useful for cinema, right? Another of the technologies that

00:21:16

exist before or after Nerf is inrea, it's called, it's this one here, it's also a paper that

00:21:23

Well, the big producers are using this technology to implement Gaussian splatting in cinema and well, here local processing power is also required on their computers and time, obviously, so here what

00:21:41

offers this type of technology is to be able to increase the quality to the maximum, obviously it is somewhat heavy for something web, as we are anticipating as users of Vista 3, but yes for other types of productions, I don't know,

00:21:54

create video, make virtual productions or do integration in Blender or 3 of Studio Max to be able to integrate 3D content on something scanned, then this would be for me the most complete of all, the most powerful of

00:22:12

everyone, what

00:22:13

It would be this software, right? And then comes, now yes, post-shot, that

00:22:17

is the one who

00:22:18

we've been talking about it all along, it's equally powerful, it's free, it's still in beta like everything else

00:22:25

We are seeing right now, this one is still in the beta process and unlike these 3 initial ones, just to take it into account, post-shot does allow us to have control of the quality of the scan or

00:22:39

of the quality of the final mesh or the final splat speaking only of Gaussian splat, then here what

00:22:46

I put in bold it's like what

00:22:48

is recommending or this direction to where we can go, I don't know if anyone wanted

00:22:54

mention something, all good, good, excellent, well, then you call to conclude this part a little, if we go, for example through Luma, Policam and Kiri, we have no problem, basically it is to process, download and then leave

00:23:19

to edit the model, that would be like the normal flow, if we go by post-shot, there are two main ways, one is to go directly to process in post-shot and the other is to try to help it post-shot

00:23:38

-shot let's say that

00:23:40

to overcome certain deficiencies that

00:23:42

has the program to be able to either cut the production time or improve the sharpness of the model, however all this is a hypothesis that

00:23:51

We have been around playing with Guillermo through some papers and videos that

00:23:57

I also leave them here, so the normal thing is to open post-shot, create the dataset there and process it on your computers, download and clean it within super splat, but well, the hypothesis that

00:24:12

there was part of a method that

00:24:14

It's called a call map where, well, post-shot mainly, let's say that

00:24:21

generates a workflow where I'm going to explain a little bit here, where it first collects the images, this is my very big brush, the first thing that

00:24:34

does is collect the images that

00:24:37

You imported it to the program, that is, all the photographs that

00:24:41

You load it into the program, that takes some time, post-shot too to be more efficient, remember that

00:24:49

post-shot is focused on the web and on total control of the mesh quality, which

00:24:56

post-shot in the second step is to be able to select only the best images that

00:25:05

you throw him into the program, or allow him to

00:25:11

take the entire dataset as from the program, then well, this would be like the first step that

00:25:20

makes a post-shot of where he is now

00:25:22

is suggested here in this example is simply because

00:25:26

post-shot makes the decision of which are the best images

00:25:29

you could use in case

00:25:32

want to take that route, one of the deficiencies that

00:25:36

has post-shot is precisely the lineup of images that

00:25:40

It's called camera tracking, it's where each photograph that

00:25:44

You took with your drone or with your camera, post-shot, what

00:25:49

What you try to do is find the position of each camera where it was taken and this post-shot process usually takes a long time, which is precisely why this call map methodology

00:26:07

I was talking about this side, what

00:26:09

does is mix this software together with post-shot to be able to optimize this timeline here, no, this that

00:26:17

This green line would be what finally

00:26:20

post-shot does is train the model with this dataset that

00:26:26

are the photographs, the position of the camera and finally well here what

00:26:33

does is train the training of what

00:26:36

would be our final model that

00:26:38

is what

00:26:38

we are looking for, so if there is a way, here is a hypothesis that

00:26:44

There is a way to cut this timeline to allocate more time to either higher quality training or simply to cut exactly what

00:26:57

It would be the production line of everything that

00:27:01

It's post-shot to just go to training, so let's say this is

00:27:07

call map methodology is what

00:27:10

It had even been implemented by production companies to have more quality and less production time, but other types of skills are required.

00:27:19

maybe right now we don't have many of those

00:27:22

We are here, no, then this will be like part of the results of an image produced with call map and then with processes with programs like post-shot, well, there can be, you can see details normally.

00:27:37

sharper or so it seems

00:27:40

the hypothesis that

00:27:40

It is had here, that

00:27:42

simply processing with post-shot, here is how the data is

00:27:47

We're going to see it now in practice, but currently, someone who

00:27:53

This is Oli, one of those who

00:27:57

We have it here in the references, I think.

00:27:59

Many of us here are already starting to understand this alin that

00:28:02

generates a lot of content for gaussian splatting among others, since it precisely realizes this timeline where this process that

00:28:13

makes call map and post-shot that

00:28:15

does it very slowly, it is constantly improving that

00:28:19

It is part of the beta that

00:28:20

You have it with post-shot, that's what

00:28:22

It's trying to be an Epic Games software, they do very well precisely that procedure of aligning images, so in addition to that

00:28:37

They do it faster, reality capture, they usually put more points, more density in their points, so that

00:28:47

the result within programs such as post-shot can be visibly sharper, than

00:28:53

the normal on the one hand, less time and on the other hand more information than

00:28:59

can give us a cleaner mesh, which

00:29:02

It usually lacks, for example, these little ghosts, I don't know if you've seen them around, these Mayans like clouds that

00:29:11

They are always contaminating our scenes, when we finish a Gaussian, the objective is to try to have something with more quality, more controlled and in less time in theory, but well that is precisely where we come up against

00:29:27

with Guillermo in an exercise that

00:29:29

is taking place, so on the one hand I don't know if everyone knows reality capture, if you don't know it, I recommend installing it, here I leave you like a small

00:29:42

guide also, it is a free software in numbers, if you are billing less than a million dollars, it is still free, if you bill from there to there then it is happy problems, finally then you can pay for that and

00:29:55

more, here we are required to have a myth of the system and here we need to install it, but well, this is still part of the hypothesis, so what

00:30:06

It is suggested here to combine these two programs: use reality capture to align or track the camera and use postshot precisely to be able to train these clouds with greater control and sharpness if necessary.

00:30:24

Yes, if we require it

00:30:25

if it's a quick project if it's a hobby project or something that

00:30:30

It is not very relevant because possibly your ideal solution would be these and you would take it in no time simply upload the images to make a click and wait to be able to achieve a result similar to what

00:30:44

we would have in postshot then well part of what

00:30:49

This flow speaks that

00:30:51

Guillermo also spoke about it, we are trying with him because it is precisely then to mix these two programs, here I leave you a video that

00:31:03

It practically brings the theory and here I leave you a blog where we can enter this blog and we go that

00:31:13

It allows me to open it, it is not letting me have a blog that

00:31:21

we that

00:31:21

It brings us the instructions with an accompanying video and the instructions on how we could achieve this procedure where reality capture is mixed for the lineation of which

00:31:33

I was just talking about this postshot, it even has a kind of code that

00:31:42

we can run it is a dot bat I don't know I think that

00:31:46

Here the best thing we will find is that

00:31:49

It is a dot bat it is an executable that

00:31:52

what

00:31:53

Basically it runs several actions on our computer like opening reality capture, loading the images, downloading the result of reality capture and opening postshot so that

00:32:04

our mesh is processed, let's say that

00:32:08

You can try this, but I also like to do it manually so I can stay

00:32:12

with two routes that

00:32:14

is photogrammetry and what

00:32:18

It is gauche in esplata, well then, well, we have here a little bit of the theory of what

00:32:25

I have been talking to you just now where the usual postshot process as I was telling you is we upload the postshot images select here in green you can see the best images that

00:32:38

We then have a postshot to the line or does the tracking camera that

00:32:43

It does it slowly but surely, let's call it that finally because a case study of this is that

00:32:51

a 30k step image process that

00:32:56

We'll see it now, because it takes us more or less 5 hours and 23 minutes if we did this process using reality capture and we don't have an error that

00:33:06

I'm going to show you here below that this could be reduced by practically 80% of the time, that is, we could have it in about 1 hour and a half more or less, no yes yes no doing it

00:33:19

the math but but you can practically save us all this time if we don't have any errors, in addition to being able to obtain a slightly better or cleaner result then well this part here from which

00:33:35

I was telling them that

00:33:36

It is the lineation that

00:33:37

It consists precisely in reconstructing this point cloud and finding the position of each photograph in space. These are two pieces of data that

00:33:47

we have to

00:33:48

Finding on the one hand the position of each photograph in space and on the other hand the point cloud are two data that can be said to be

00:33:59

process in reality capture precisely to solve that problem that

00:34:07

It is the lineation and the improvement or a cloud of points more denza if there with Guillermo in this exercise what

00:34:17

what happened is that

00:34:18

Well Guillermo uploaded around 3000 images but in this reality capture procedure what

00:34:25

what happened is that

00:34:26

In the line as Guillermo had four data sets, one from the drone or maybe several from the drone taken at different times with different techniques for the facade, for the enital and for the orbits, it could have happened that

00:34:41

Some of these frames or images might have an alignment problem, just like what happened to many of you here with Pete, and when you want to sew a panoramic view,

00:34:52

have a very smooth ceiling or a very smooth wall that

00:34:56

They do not have control marks, so the same thing happens right here, we would have to

00:35:00

manually in case

00:35:03

will not separate models instead of having a single model align with control points manually in the program where we can tell it that

00:35:14

This photograph is the same as

00:35:16

This or a certain point of the photograph coincides in this way, let's say that

00:35:25

join those who

00:35:27

which

00:35:27

Here in reality capture they call them components that

00:35:30

They are practically models, in this case it generated four components from a data set of images instead of generating them only and with these four components you would have to

00:35:43

align them manually then well here again oly here I put a video in advance so that

00:35:51

see a little bit of what

00:35:52

he experienced that

00:35:54

It is precisely with this reality capture procedure that you save or can save a lot of processing time without improving the quality of the image, however, if you encounter this problem that

00:36:12

You have several captures at different times or with different cameras, it could be that

00:36:19

the uprising that

00:36:21

you made images have some alignment error this would suggest that

00:36:26

you will have to

00:36:27

align it manually with control points made in reality capture as it happens here and in his case and mine and surely all of you it would take too much time to learn and do it manually then we would be

00:36:43

getting a little to the same point where we saved time on the procedure with post shot but we invested man time trying to align these components to be able to sew them and now we can take this position of

00:37:01

cameras and nine points to post shot to advance the process well in short what

00:37:08

Kioli also suggests that

00:37:09

tells us that

00:37:10

If we suddenly start taking practice this is a let's say that

00:37:16

It takes courage to use this reality capture methodology but if not, we still prefer that

00:37:24

the machine then works well until the conclusion what

00:37:27

says is that

00:37:28

Well, let's sell the dataset to post shot and that

00:37:32

Let's think about the capture for another occasion then because they are paths that

00:37:38

Although using reality capture and practicing can lead us to reduce times and improve quality, having this type of errors can also lead us to the conclusion that it says that

00:38:00

Well post shot keeps improving as it continues in beta and the improvement that

00:38:05

What they intend to do in the next quarters of the year is precisely to improve the alignment of the cameras so that

00:38:14

It is a faster process, in fact there is already a paper, so I think that

00:38:17

everything goes to the line like that, no point and then Mario, I don't know if what

00:38:21

I have also experienced doing a reality capture process, mixing the two makes the model always aligned with the horizon, but in post-shot, although

00:38:33

then you can move it but it puts it tilted where it has fallen is the other difference that

00:38:38

I have also seen that

00:38:40

It's fair in reality and capture boom you always leave it aligned to the ground I don't know if it will take GPS from the drone or I have no idea but the models that

00:38:50

I go from reality capture to post shot and they are perfectly aligned on the horizon or on the ground in the shot, but if I leave it working in post shot it puts it as it falls and then you

00:39:03

you have to

00:39:03

reposition if correct but well that part can be resolved very quickly aligned but if it is a good observation then chi well from here let's say that

00:39:18

of this flow that

00:39:20

As I was saying, here I leave the instructions on this blog and right now we are trying to do something through the reality capture methodology, what

00:39:31

we obtain is precisely to export a CCV that

00:39:34

It's a carbajal there they told us some point with ptg that

00:39:45

we could copy those coordinates into ptg so that

00:39:49

I would align it faster and then everything evolved but for now it is something similar and the play was done in reality capture that

00:39:57

It is nothing more than

00:39:57

the point cloud not the Gaussian not a model but simply the point cloud the combination of the CCV the p ley that

00:40:07

Here we have it and the photo dataset is what

00:40:11

we could drag here in post shot so that we could now process only what

00:40:17

It would be the Gaussian splatting model, so if, for example, we are lucky enough not to have this error that

00:40:25

do not separate the model by components this would be a great workflow from reality capture to post shot in case

00:40:34

We have this little problem, well, we will have to

00:40:37

simply accept the fact that

00:40:39

we will have to

00:40:40

work a little more to be able to align them, take practice and occasionally, well, we have, like,

00:40:49

a flow of our own or a more established flow, knowing how long it takes us and what

00:40:55

We look for when we choose this autology of position and in reality with the study it is to give a job as easy as possible but because they believe that

00:41:07

we would take fluma policam or Kiri and it is not that

00:41:10

They may not be, but there are other faster routes, so I'm going to need to face the challenge, as the learning curve of aligning with reality capture has said.

00:41:25

is what

00:41:25

I see, it seems a little complicated to me and it might take you out of this flow and you say, look, post shot, everything takes up the machine, what

00:41:34

you need and that's it, look to see what

00:41:37

reality capture is missing so that

00:41:40

do it all in one component maybe you need to rename the image sequence because

00:41:47

I specifically did not use several cameras, but if there were several runs, then there may be a lack of continuity between one pass and another or the nomenclature of the files. I have a program that

00:42:02

It renames these very quickly and as I did everything I made several video clips and I made several exports from pocho to have the frames what

00:42:13

yes that

00:42:13

Then I went and some frames were called the same, so with the renamer I put in front of not all the files, a whole batch has the same, so I go and say okay when I finish all that

00:42:26

It was the frontal scan of the facade to throw a few at 45 degrees that

00:42:34

They join me with the garden before putting the garden zenith, maybe it is necessary

00:42:39

prepare the material in a way that captures reality so that

00:42:43

No, don't let the components destroy it, always taking into account that

00:42:49

well any any this uprising that

00:42:52

You do a scan, for example here, if you take, for example, images of the facade here, let's see that

00:43:03

This is a facade, this whole part here is if they take images of the facade in this sense, well always capturing towards the facade in some way when we change to scan the roof it always has to

00:43:18

there must be a significant overlap, that is, all of these photos of the facade or a large part of the photo of the facade have to

00:43:25

see points that

00:43:27

in turn can be seen from the 45 degree shots taken from above so that

00:43:33

Well, the software allows you to line them up and it's just as simple or something that

00:43:42

It just occurred to me with my specific building if I do the zenith of the garden and start on the left side and finish on the right there with the inclination of 30-45

00:43:58

degrees and return from right to left so that

00:44:02

In the last photo of the zenithal shot, the one with the first tilted at an angle could be a part and not return to the left side and do the scan again in the same direction

00:44:16

but

00:44:16

going back to the counter plays a clear question if it is a possibility if we go through that

00:44:25

also good gauche is the scan that

00:44:28

It is taken here if they have a lot of the basics of photography, not exactly the overlap is key where one of these photos sees 60 or 70 photos of the facade and suddenly they have one

00:44:47

from above because part of these facade photos will have to

00:44:50

see 60-70 percent of what

00:44:52

see the ones above or vice versa so that

00:44:54

because the program can understand where photos from one camera join with those from the other in the order in which

00:45:04

doing the same also influences, that's what

00:45:06

He referred to that

00:45:07

Here you are looking from the front of the church or the house to the right at 90 degrees on the slope, but then you scan the roof at 45 degrees from the

00:45:20

last one on the right, no, from the very right, not from the front, I say this because

00:45:26

Likewise, if this goes from here to there and then from here start in that direction to have that same order, I would investigate there because

00:45:34

I think that

00:45:34

Reality Capture is 100%, it is very powerful, but it is also very snobbish when it comes to asking for it and I want to confirm if they take into account the order of the files

00:45:44

If it is a possibility, I would think that

00:45:47

if it helps you because

00:45:49

There is photometry software that

00:45:52

They tell you, like for example, Policam asks you, the images that

00:45:57

I am uploading them in the correct order, that is, they have a sequence and you can check the box that

00:46:04

yes or what

00:46:05

no and I suppose that

00:46:07

Policam makes configuration decisions to be able to process what

00:46:12

you say it makes sense to me that

00:46:13

could be a factor that

00:46:14

influence that

00:46:16

this can turn out better, not that

00:46:17

you can't, but simply that

00:46:19

can come out better, faster, right? Okay, well, so as not to dwell too much on this, I say that

00:46:26

Here I leave you this process, the procedure, now we will try to emulate some of this to see if it is a question it seems to me

00:46:35

Yes, go ahead, Pietro.

00:46:37

No, Guillermo, when will you name the files, accordingly? The southern ones, I don't have them for a moment, but I say, but yes

00:46:47

They listen to me

00:46:50

Yes, I didn't understand the last thing.

00:46:52

You said that

00:46:53

I'll file your name, right? Yes, I do,

00:46:56

I, the technique that

00:46:58

I used, obviously I had to

00:47:00

It was video, Oscar went to photo and I went to video, because,

00:47:04

Well, in the end, with video, you scan faster than

00:47:06

with what

00:47:07

taking pictures

00:47:08

So, using the pochot technique to be able to convert to frames and on top of that select the best frames, because

00:47:18

If, for example, you have had a blur,

00:47:20

a ray of light that

00:47:20

It got into your lens, your drone shook, pochot, that's a frame, it doesn't take them into account

00:47:27

And then, I use pochot to export the frame sequence, reality capture, to align the cameras and then pochot again to process

00:47:35

So, as there are several files, for example, I make the first part of the facade perpendicular and hit stop

00:47:45

That's a file, that's a sequence of images

00:47:50

Then I make the other facade, another video file, another sequence of images

00:47:55

Of course, when I export them, the program can name it facade 1, and then I, if I can name it facade 2, then, each sequence or each group of images, depending on the recording file, I have it named

00:48:11

in a way

00:48:12

But they all have, that is, the first facade or the first video file, would be facade 1, 0, 0, 1, 0, 0, 2, 0, 0, 3, 0, 0, 0, 4

00:48:21

The other facade would be facade 2, 0, 0, 0, 0, 1, 0, 0, 0, 2

00:48:26

And then each file sequence has a common nexus of the name

00:48:31

But, later, you understand

00:48:34

Yes, that should be able to help you, William, but I think that

00:48:38

I don't know, I'm with, I'm with Pietro, who

00:48:40

the doubt that

00:48:41

has

00:48:41

Ah, but go ahead, yes

00:48:43

The question, the first one, then you upload each file, he will elaborate, he will put together, the next one, or he will give you everything

00:48:56

Everything, everything, everything

00:48:57

In fact, what

00:48:58

the bat file said

00:48:59

of the tutorial is a script, which

00:49:02

You double-clip it, you leave it prepared as the tutorial says, which

00:49:05

one has to

00:49:05

create a folder that

00:49:06

It's called images, and the script does it alone

00:49:11

It opens the reality capture by itself and aligns itself

00:49:15

So, in that image folder is where I have put all the camera settings, that is, the original photos, all the scanning that

00:49:26

you've already done it, throw it in that folder

00:49:28

And then, there is reality capture line up

00:49:31

And the second question is this, this

00:49:35

True, really, it is important to take the shot as you say, in sequence, when in the end the orientation of the cameras is done with the control points

00:49:50

No, it's that

00:49:51

We don't have the same dialogue there, about the checkpoints

00:49:54

The thing is

00:49:54

the reality capture in front of Pocho, that is, at first we did everything with Pocho

00:50:00

I am referring to the setting sun

00:50:03

It's just that

00:50:04

is recognized

00:50:05

It is a confusing term that

00:50:07

maybe Talia or Mario

00:50:09

Me because

00:50:09

I know what

00:50:10

He is talking, but it is not because of the sun's blindness

00:50:13

The checkpoints that

00:50:15

He said, they are when reality capture separates the scans into different components

00:50:21

Look, show the components crash, Mario.

00:50:25

Yes, of course, not here.

00:50:27

It's just that

00:50:28

There is, the way to interpret the reality capture scans, all the images that

00:50:32

You have put the scanning into it, it, due to its internal functioning, has made the components

00:50:38

So, from reality capture to pocho, not all the scans work for you.

00:50:45

For example, the roof, the zenithal garden, separated me from the facade at 30 degrees.

00:50:53

And then, either I export one thing or I export another.

00:50:56

And checkpoints are for merging those two models, not the sequence views.

00:51:02

I don't know if I explain myself

00:51:05

I don't know if I explain myself

00:51:06

How about a checkpoint? I understand that

00:51:08

it

00:51:10

The software too

00:51:11

Here we are seeing the components, this is the facade, here we will soon have the interior, correct, or the roof

00:51:21

Why did he do this? Well, because

00:51:23

reality capture did not understand at some point how the images were joined together

00:51:30

were recorded with the drone and those that

00:51:32

were recorded for the facade that

00:51:34

It was possibly another drone or the same one, right? Or if they did it at different times

00:51:38

So that's a possibility that

00:51:41

has reality capture of that

00:51:44

This can happen, right? And when you put this in post-shot, it can also happen

00:51:50

It has been decided nothing more than

00:51:51

It simply doesn't show up like reality capture does, but rather

00:51:55

post-shot, which

00:51:55

does is that

00:51:56

take the best pictures

00:51:58

If you put in 3,000

00:52:00

images, you have the possibility that

00:52:02

Take only the best, here I will sell some, or you can tell them that

00:52:09

take all, completely all the images, right? Then

00:52:12

I haven't tried that yet.

00:52:14

be all of them, I always let him choose

00:52:17

I'm going to get in here to put the one

00:52:19

I'm putting it in, but it's a video

00:52:21

It is not

00:52:22

Here's what

00:52:22

tells us post-shot when I track a video or image dataset is that

00:52:27

I have 2475 images in this video that

00:52:31

I just imported, of which post-shot, here it is limiting me to 166

00:52:38

Why? Because

00:52:39

has this value that

00:52:40

says to use the best images and allows us to limit it so that

00:52:44

be a little lighter

00:52:45

If I tell it to use all the datasets, then I'm going to use the full video with 2475, which is

00:52:52

suggests that

00:52:53

It's going to take a lot longer, and I might find bugs like Reality Capture did.

00:52:58

But in this way, using the best images that

00:53:02

I can tell you, because you know that

00:53:04

50 percent are not in 1250 images

00:53:08

Okay, they let you, right? I hadn't tried that.

00:53:11

Dog

00:53:12

So what

00:53:13

post-shot is from that video that

00:53:15

I put that

00:53:16

select the 1200 best images for you

00:53:19

can be aligned that

00:53:21

have that validation

00:53:22

But Mario, here, as I have done with this tutorial, I have exported the best frames of each clip with post-shot and I have taken it to the reality capture script, and it has still worked for me.

00:53:37

made that mistake

00:53:38

So it doesn't influence 100 percent of the best images

00:53:42

Right? What's like paroptera? Time, right? Of course, I've made the filter that,

00:53:52

and I have already made the mistake of separating it into components

00:53:56

What is it when I decided you? What's weird? I only have the roof, and when you told me, I went to the second layer of components, and Ibi, the facade.

00:54:05

When suddenly I can't be, if I have thrown the photos of the facade at him and he has blown them, and it was clear, that

00:54:11

He did it perfectly, but he did it in another component.

00:54:15

Correct? Yes, yes, in fact, here too in the exercise that

00:54:20

This person, Oli, presents at the end what

00:54:23

You decide is, well, to put all the images into reality capture just to get all the value, no, it is to put the complete dataset

00:54:31

Why? Because

00:54:32

Post shot would take a long time to process all these images when Reality Capture can do it.

00:54:38

This, in the end, is saving time and having more resolution in the point cloud than

00:54:44

genera reality capture

00:54:46

On the contrary, Mario, if you look if you go to the post shot tutorial with reality capture, it cancels and that

00:54:52

we have the post shot window

00:54:55

I was reading, there is a point where

00:54:58

It takes you like, let's see, no, up, up to the top, there it tells you, let's see, go down a little, here, click on reality capture to the link that

00:55:08

It puts you down there, there, setting, that

00:55:11

Here somewhere, I have read that

00:55:14

tells you that

00:55:15

use the third of photos that

00:55:17

have

00:55:18

That is, here they are given the sun protection thing

00:55:21

I read below that

00:55:22

It gives me more down

00:55:24

I think that

00:55:25

does something to you that

00:55:26

I do, that

00:55:27

is to select a column in Windows, it gives me more down

00:55:30

Yeah

00:55:32

How did you do it before to be able to quickly select a portion of images? And then there, there, this

00:55:39

So, I say, but then you no longer need the entire dataset from post shot

00:55:44

So no, I'm going crazy with this part.

00:55:47

I don't know if it is necessary

00:55:48

give him a lot of images or do you have to

00:55:50

give him the minimum so that

00:55:53

is oriented

00:55:54

Correct

00:55:56

Well yes, we're talking about this in general, yes, exactly about this, right? How many images to choose?

00:56:02

There are two options here, well, three.

00:56:06

It's one, we put all the images and that

00:56:08

post shot work with all

00:56:11

Two, we allow post shot to choose the best ones by limiting it in the software.

00:56:21

And three, as it was done before, right? That what

00:56:23

Guillermo says, maybe I take this row, this row, this row, randomly, and you dream that

00:56:29

I care to be able to reconstruct my which is random, finally, right? In the exercises, these of Oli, the same person who

00:56:39

We talked, he made a video where he mentioned how many is the minimum to be able to reconstruct a model

00:56:45

That is, a one-piece model, a chair or anything else

00:56:50

And it goes to the limit down, where well, precisely with up to 13 images it seems to me, this one can reconstruct a piece, right? What

00:57:00

suggests that

00:57:02

Well, perhaps the entire dataset is not necessary, because

00:57:06

We have a training process with AI when reconstructing the model, since

00:57:12

Well, it may be, let's say, that

00:57:16

interpolate, right? Between one image and another, randomly, that

00:57:20

In this case, the post-shot helps us with that process, right? So, well, to not overwhelm you with concepts of this type, here in the conclusion, you can review it, well, exactly if

00:57:36

They want to use the Reality Capture method, yes it can bring us many improvements in our workflow and improvements in quality, but it suggests acquiring new skills to be able to cook these components that

00:57:51

they can be small

00:57:52

errors, as it doesn't usually happen in PTG, I think

00:57:55

Most of us have some experience with that, right? If we don't want to

00:57:59

deal with that, then, to work the whole process with post-shot, which will undoubtedly be better than

00:58:11

any of these other three processes, not more efficient, but better in terms of the quality of the final model

00:58:19

I don't know, here I leave you this information and well, maybe, I can be very better, maybe I have confused something and you ask me some questions in this regard, but in summary, my conclusion would be,

00:58:35

Well, take these options when you don't have any requirement

00:58:40

very strict about the professionalism of the final model, take post-shot directly with the entire dataset inserted there, in such a way that

00:58:49

you can play with, come on, the amount of images that

00:58:56

they want that

00:58:57

with which

00:58:59

they want that

00:58:59

program to work, but this is going to take time, right? The last scan that

00:59:03

I made it took me 23 hours processing in this program and well it is a time that

00:59:08

it would be necessary

00:59:09

assume, right? Depending on the quality that

00:59:11

they are looking for

00:59:12

How is the quality of the post-shot measured? Well, on the one hand, the number of images that

00:59:18

aline and, on the other hand, the quality of the model that

00:59:22

These are the case steps that

00:59:24

He was also talking about it a little while ago, right? Usually, Guillermo mentioned here that

00:59:30

one has to

00:59:30

take a portion of these images, right? Not all of them, that

00:59:34

It would be excessive to take them

00:59:36

Maybe it should be

00:59:38

Identify what that percentage is, if it is half of the images, if it is a third, if it is a maximum, for example, a thousand, and do tests.

00:59:47

And on the other hand, this is to generate the point cloud

00:59:51

And, on the other hand, to properly generate the model that

00:59:54

These are the splats with which

00:59:56

We basically play at the end and that is the internment

01:00:01

the program, because these are the k-steps that

01:00:05

We have here, the basics are 30 and here the program puts us under pressure, well, tests that

01:00:13

It is holy, that

01:00:14

more look, minimum gynecologists at 30 steps, 5 hours 23 with a 2070, 3070, sorry, a 3070 card

01:00:28

So what do those 30 acas refer to that

01:00:31

They speak? Well, it is precisely also a possibility that

01:00:35

has the ability to clean these little clouds that

01:00:38

They have it here and that

01:00:40

The model has greater sharpness at the time of completion. This is actually talking about the model, right? So, the higher that number, the less waste it generates? Less waste and more sharpness it generates.

01:00:55

So, just the fact that

01:00:58

Saving time in this part would allow us to grow this so we could better use that time to have higher resolution in the mesh, which

01:01:10

is properly what

01:01:12

We're playing in 3rd sight, aren't we? Here I have an exercise that

01:01:16

I'm going to show it

01:01:18

And a question before opening the program, when I did, well there the other caesteps, made another video, please, there to the potso

01:01:26

Yes, I'll be here for another video.

01:01:28

There, these who

01:01:29

They put medium, the max is platformed, in the middle, that's my difference between that one and the other, stop training after

01:01:39

Yes, here I say finish that

01:01:40

There are not many terms, there will be

01:01:42

go into a little depth with this, but well, this whole part up here is what

01:01:48

It is supplemented in reality capture, that is, there are two concepts, the position of the cameras, they are the cameras and the point cloud

01:01:59

And this part down here is the actual construction of the model, that is, here we are talking about the splatin gauján, so here we have two options, do it in postshot, it takes more machine time, but maybe

01:02:19

much less time man, or use it with reality capture to make this part more efficient, obviously with the risk or the need to have your skills or those abilities in reality capture, to finally be able to, well

01:02:35

here, if we have no other choice but to

01:02:37

train in postshot and have this final model, then in relation to the model, here is the only value with which

01:02:45

We can play, I've played here since I was 30.

01:02:49

It's the basics, I've put up to 1000 or 5000, I've left it there to test how it works for me, however, I wanted

01:02:56

show them what this looks like, no practice

01:02:58

Hey, one thing, I don't know if you remember Mario that

01:03:04

come down, he once spoke to us, because

01:03:06

It is going to become popular, that

01:03:09

If I don't remember more, I remember less than

01:03:10

Someone correct me, that software ends in the camera positions, I wonder if using it, not popularizing it, can use it, help with postshot or reality capture, already selling it from the beginning without training

01:03:30

Correct, sounds good, Marco, however, here you go a little bit into this, whoever likes it, you will discover that

01:03:41

Two pieces of information are required to be able to use the information that

01:03:45

It comes from a program like reality capture for postshot

01:03:49

The first piece of information, as I was saying, is the position of the cameras, which

01:03:52

This is what

01:03:53

You suggest with this little program, is to find the position of the cameras and their coordinates, but then we need the point cloud, which

01:04:03

These are the little dots, which

01:04:04

It is already the triangulation of key points, where in this photo, for example, photo 1 and photo 2, there are points that

01:04:14

They see it right there, the best one is a spot, photo 1 is seeing it and 2 too, then a triangulation is generated and then we get this point cloud, maybe with this program that

01:04:25

suggests that the framework can solve this and this, but we would stay

01:04:29

still with this, right? Well, let's say that

01:04:34

the problem of the point cloud as such

01:04:38

So, yes, I say, to that

01:04:40

test, precisely, and remember, I say, this whole part here is precisely to solve a problem that

01:04:47

exists right now, which maybe tomorrow postshot will show that

01:04:50

version 2.0

01:04:51

and have a new alignment mechanism, because

01:04:56

Well, it finally makes us forget about reality and capture the moment, right? That's another possibility.

01:05:02

I mean papi huiza

01:05:06

Very well, I don't know if, I hope so.

01:05:09

Mario you are not listening to Pietro because

01:05:12

wants that

01:05:13

doesn't listen to him

01:05:17

It's not a papi huiza and I used it differently

01:05:24

You create it

01:05:25

I don't hear him but well, if he tells me

01:05:28

OK

01:05:30

Lo papi huiza, you create it by saying for each program, each column eraia in which the photograph exists

01:05:41

I imagine that

01:05:42

pair for 100 300 more photographs is too much of a struggle

01:05:47

Okay, hush, everything's fine there.

01:05:49

Yes, he's talking to Pietro about that.

01:05:51

Ah, I tell you

01:05:52

Yes, understood

01:05:54

Yes, that

01:05:54

you think that

01:05:55

It's too much work for Papi Huizar, isn't it? Because

01:05:58

It's you, who

01:05:59

create the file to satel

01:06:02

They ask me, sorry

01:06:05

No, no, no, you are that

01:06:06

He's talking to Pietro about that.

01:06:08

Very good

01:06:09

The daddy runs away, before he comes out, what do you think the software program tells him how you took the image? Imagine the struggle that

01:06:17

you must do to enter all the data of the photograph that

01:06:23

you took, because

01:06:24

In the 360 panorama you only put the column and the line

01:06:29

And now, how much raia is how much column

01:06:33

But here is a different case

01:06:36

Yes, it has to

01:06:37

be right, that

01:06:38

I tell the truth

01:06:39

I've been looking like Papi Huizar for a long time, but if he's right, the image that I've been looking for is no longer reaching there.

01:06:45

If you put a column and all that, it hasn't been useful for this, for nothing.

01:06:50

Unca, ready Mario

01:06:53

Ok, okay

01:06:56

So, Papi Huizar is left

01:06:59

ruled out, right? What

01:07:01

understood

01:07:01

OK

01:07:04

So, I go here

01:07:05

When ideas come out,

01:07:06

suddenly an idea or a crazy thing, suddenly it is testing and a part of the process, it can be avoided

01:07:14

Yes of course

01:07:16

Yes, always, it's always possible, isn't it? That part

01:07:19

And the other one doesn't say crazy things, right? No, it wasn't Marco.

01:07:26

Crazy Marco

01:07:28

Well, I'm going to show you here, this is a scan of a 3D model.

01:07:36

This is not made with photos, but it is made with renders, but for illustrative purposes, I like to show you this because

01:07:44

It looks organized, right? This is made in 3D Studio Max, the model exists, or in Blender, so you could do it

01:07:51

This model is made in 3D and here is the inverse

01:07:56

From a model that

01:07:58

It has a Mesh, I generated a Gaussian Splatting with the goal of later merging it with the real terrain

01:08:06

But, well, let's forget that part about that

01:08:08

It's 3D and let's go and check it out properly, like it was captured less than a year ago, right? So, one of the things that

01:08:19

I want to show you here, well, it is that

01:08:20

Here is the capture, having all the freedom, in Blender, a 3D Studio Max, to take photos from wherever I want, practically with a rule

01:08:32

Still, for the sake of time and all that, I decided to generate a couple of orbits.

01:08:36

Let us remember that

01:08:37

for Gaussian Splatting, then, the recommendations in all the papers that

01:08:43

I've seen, it's part of just generating orbits, right? These orbits, what

01:08:51

They also suggest that

01:08:52

From them, well, we will be able to have the necessary information to align this photo with the one below.

01:09:00

And, rather, in a complementary manner, what

01:09:03

They suggest, therefore, being able to have filler images, which

01:09:07

In this case, there were very few, these are the ones above, to be able to obtain information from the roofs or the ceilings or the entire upper part, right? So, to the best Guillermo, in your case,

01:09:20

something that

01:09:22

It occurs to me based on the evidence that

01:09:25

I have done, it is, therefore, yes or yes, always, to have orbits at different heights

01:09:30

And, from that, you can complement it with overhead photographs at 45, 0 or whatever angles you want.

01:09:40

you need or time allows

01:09:43

And, in addition to that, well, also being able to have captures of the facades, either at floor level or at level, precisely with Drone, but these orbits could be, could be that, that way to be able to

01:09:59

linking a zenithal aerial session with the facades, that is, that transition that

01:10:06

can allow us to have the best of both, both techniques, right? There, over there, vitos scans and they have to

01:10:13

much that

01:10:14

see with photogrammetry techniques, which is, which work very well for this method as well

01:10:20

But how do you know where I also fixed the problem, the tip that

01:10:25

You gave me from heaven, that

01:10:27

little to generate the sky from the last pixels of the scanned horizon, and in the first, the first scan that

01:10:35

I did not take out

01:10:36

the horizon and it came out brown like my giste of the mountains and all this

01:10:41

But then, since it wasn't a terrain like this, when scanning in orbit, I ran into the problem of how to capture the horizon from the different perspectives, right? So, I took facade one and

01:10:59

I ran parallel, but oriented to the right

01:11:04

It's like 45 degrees down, 45 degrees out and moving parallel to the facade

01:11:13

For me it was a very strange thing, but it was the way to be able to have the sky swept in the background with the different parallaxes, as I move parallel to the facade, then return along the same

01:11:28

facade, but against, minus 45 so to speak and then the other facade and the other facade

01:11:34

So, when it came to taking the sky, I also had a problem.

01:11:42

I understand a little bit of what

01:11:44

I have to

01:11:45

Speaking here, Guillermo, this aerial session was done there on a piece of land

01:11:50

Here we saw that

01:11:51

It was practically one of the first and there were a couple of orbits, practically at the same height

01:11:58

Or I don't know if you made this frame or me, I don't remember.

01:12:02

Well, here's what

01:12:03

The goal was precisely to obtain this land in order to build a building there.

01:12:09

Finally, being able to have a 3D model, right? And what

01:12:12

Now Guillermo is now of this effect that

01:12:15

It is happening here where in post shot we can tell the program when processing that

01:12:23

He built or completed a false sky for us, that is to say,

01:12:27

based on the information that

01:12:29

we have here on the perimeter of the model upwards, I can build what

01:12:35

It would be a sphere that

01:12:37

It's like the hemisphere or the 360 of the planet, right? We're having in this 3D model something like this, right? So here we could have the complete sphere where taking that last color of the complete pixel

01:12:53

what

01:12:54

the sky is blue and taking the pixel of the floor limits and complete that

01:13:00

The floor is grey or brown, isn't it? Today I'll tell you what

01:13:03

Guillermo mentions that

01:13:04

He took the pixel of the brown mountains and added more rum to it, right? Right, that's right.

01:13:14

So here what I

01:13:16

already testing with Vista 3 I also discovered that

01:13:20

maybe it is not such a good idea to complement the entire sky, but

01:13:26

In view 3 we can manually take the color of the horizon or the sky and in view 3 put it in that color to complement it

01:13:36

Did you leave us 3 views in Gaussen esplatin? Now I'll show you my exercise, right? What?

01:13:43

What do you mean by this? Mario, and the possibility of a 360-degree view of a 360-degree photograph replacing that space? Yes, still, 3D views are like HDR, yes, that's also a possibility, right? What exactly is that?

01:13:57

Now we take the color of the sky and put it as a background at 360 degrees or we can put a background photograph at 360 degrees.

01:14:09

That's also a possibility, isn't it? So here, well, in this image we would have to

01:14:14

finish cleaning everything

01:14:18

we look here so we can stay

01:14:20

only with what

01:14:21

It is necessary to limit the camera so that we cannot move too much because

01:14:26

we're going to start making noise, otherwise we'll stay

01:14:29

in the place that

01:14:31

We are interested in that

01:14:32

This is the best terrain, isn't it? And from here on, it's good to limit it, isn't it? But what

01:14:38

Guillermo mentioned that it has to be

01:14:39

See with this, there will be times when

01:14:42

works for us and there will be times when

01:14:44

As the lion says, well, we can delete the sky and put a 360 in 3-point view or a color, right? Mario, I haven't even done it, but it just occurred to me that

01:14:55

You could make a fair, let's say 3D, and fill it with a 360 image, so that the surface

01:15:06

In the end the model would be with that limit, let's say

01:15:11

Yes, it is a possibility to mix them here, you mean mixing a model 13 of the sky with a model, you mean? Of course, like with a wait that

01:15:18

had a 360 photo inside so that,

01:15:22

In the background, infinity, let's say, the background was a sky made apart

01:15:29

Yes, that is a possibility.

01:15:30

Yes, I'm going to think about it there, but well, I think that

01:15:34

which

01:15:34

we talk about here is that

01:15:35

Instead of having a black background, we could take, I don't know, the horizon of the sky, let's think that

01:15:42

the horizon is this color and see how it works, right? This is a possibility 1, which

01:15:48

It is bad light and it is the one that

01:15:49

I would try to take

01:15:50

It even helps me to see errors and upload a 360 here, which

01:15:59

is to open the panorama, which

01:16:03

Maybe it was taken with the drone at a certain height and it could allow us to merge a little better or maybe in another creative way, what

01:16:15

would be our model with that photograph, right? See here we have the panorama, let me limit the camera so that

01:16:23

You can visualize it, there it is, right? We would have something like this, taken from a certain height, these are possibilities, right? Being able to build this faithfully in Gaussian Splatting, adding a color or having this spherical photograph, right? That I can make it spliced

01:16:43

With our model, when you zoom in, does it zoom in? Yes, I have it limited.

01:16:50

No, no, but you see the background that

01:16:51

does not respond

01:16:53

Ah, no, of course, this photograph was taken on the 30th floor, yes, but it's that

01:16:58

This technique also occurs to me and it has limitations that

01:17:01

not interacting with the background, then you only oppose a strange one, Leon, sorry, let's have the 3D sphere, there it would be interactive, but they are creative solutions, right? That we could have there in a new case

01:17:24

Or when they put in the Gaussian option, things that

01:17:28

has in the 3D game 3D view, which

01:17:30

It is the environment that is, that

01:17:32

I think that

01:17:34

It's more worked on the other side, I think.

01:17:37

Ok, they are possibilities, right? Also the same to attend to like this of what

01:17:43

I was talking about, right? Speaking of case steps, I'm going to close this, also when rebuilding this model, let's think that

01:17:53

I brought this from Reality Capture or did it directly here in post-shot, right? This image is generated with 30 case steps. These steps, see the difference. Now I'm going to open a curtain 400, right?

01:18:09

In this model, since it has this type of resolution, I don't know if you're tired of appreciating it, it looks a little blurry, especially here in the access

01:18:19

For what

01:18:20

I need, that

01:18:21

Seeing it at this distance is enough, it's like the printing resolution for a billboard. If I need more resolution or to see it a little sharper, then I could work with a larger point cloud.

01:18:34

dense and upload it to the training in post shot, or just upload it to this, right? I'm going to open the 400 image here, this model that

01:18:46

I'm going to open it right now, it's 400 case steps. If anyone has any questions about where that fits in, I'll project it here right now, right? It's the same information, the same cameras, but in this case I left it longer.

01:19:00

training, with this factor, then if I open one image and another, we can see, I don't know if because of how I share the screen, this could be more visible, how one has more density than

01:19:20

another, that

01:19:21

another you can notice the difference, yes, I imagine that

01:19:24

the transmission takes something away from us, but if we see a difference between one and the other, so this is good, this is solved, because right here no, or more than

01:19:35

solve it, because that's what it is

01:19:36

You have the courage to use post-shot to be able to control precisely this information, right? What is this point, right? One is with 30, the other looks with 400, but if you are willing to wait

01:19:48

more time to process, they could do it with much more, right? To have more clarity within what

01:19:56

Is this post-shot? First, the base, I'm going to ask without closing that, I ask again if you do it all in post-shot, the 3,000

01:20:07

s in max splat count, what does it also influence in this sharpness or only from below? Yes, it seems to me, I say, we will have to

01:20:16

check it here in the help section, but it seems to me that

01:20:20

that part is focused on the point cloud, precisely on the points, so possibly here we can control if what

01:20:30

does reality capture well with the part of giving us denser meshes, I don't know, it's just that

01:20:38

I touch or that one and with that one is where the processing times change for me

01:20:42

Yeah, another thing that

01:20:43

You can change the timings here. Well, on the one hand, it's the number of images, okay? On the other hand, if you look here, my images are probably captured in 4K with Midron, right? But post-shot

01:20:58

you are suggesting to me that

01:20:59

reduce the quality by 1,600

01:21:01

pixels instead of using full, that is, complete with the full resolution

01:21:04

I already adapted it there, I adapted it so that

01:21:07

whether they were in source form or adapted to 1,600

01:21:10

so that

01:21:11

No, don't burden him with more work for that reason.

01:21:13

It's possibly a solution, right? So, if I want to use them in 4K, well, here I remove this function and it will take all the images from Midron in the maximum resolution that

01:21:27

It could be 4K or the original, right? So, I can help you optimize things here, this post shot, by lowering the resolution and the amount of information in this.

01:21:38

This last one, well as you say, I also have my doubts about this one, it would be necessary

01:21:41

research here in Help to see this one

01:21:44

It refers to, but I want to think that

01:21:46

like what

01:21:47

We are supplying with Reality Capture is on the one hand the camera alignment, and on the other hand, the point cloud, which

01:21:53

The denser and more even the point cloud, the better our splat will be.

01:21:58

So, this possibly going up here, generates a denser point cloud, but let's remember that

01:22:05

which

01:22:05

I what

01:22:06

worse post shot, because that's exactly what it takes, right? So, for example, I had played at putting it at 1,500,

01:22:18

2 000,

01:22:19

I think that

01:22:19

It has been the last one, that

01:22:20

has been the one who

01:22:20

It made me happier, but I hadn't touched the one below

01:22:24

So, or if I had raised it to 60, but not to 400, then I'm going to do the counter, I'm going to put Max splat at 1000, which

01:22:32

I know that

01:22:32

It has gone lightly for me and I'm going to put 400 below, let's see, that imbalance against it, let's see what result it gives me

01:22:41

Yes, look, right there what

01:22:42

I think, let's say that

01:22:45

of that value is precisely this that

01:22:47

you're seeing here, no, this is like something similar to what

01:22:51

It happens with the point cloud, here we have a point cloud that

01:22:54

Maybe it's not very dense, by increasing that value we could have the densest, as happens with Real Eticaptor, and therefore, obviously have a better reconstructed splat, right? But they are small.

01:23:09

values that

01:23:09

maybe, if we think that

01:23:12

It's going to be 3rd sight, I've realized that

01:23:14

3rd view compresses, see here the level of compression, that's it, well I say, it's still very bad, I don't know what's happening, and then my next test here is going to be today is

01:23:25

precisely with this model at 400 case steps to see, well, precisely if it improves me, right? Something that

01:23:32

says 3 of view that

01:23:33

we have to

01:23:34

take care away from the weight of our model, as it is precisely the number of gaushans that

01:23:40

we have, they mention that

01:23:43

a million is like that

01:23:44

the most that

01:23:45

we should be allowed to have an interactive that

01:23:49

is fluid, but if I import my 400 version here, it gives me around 3 million, which

01:24:01

I still don't understand why, but Vista 3 compresses it, so, yeah maybe I have to

01:24:08

see with this dynamic resolution, maybe it has to

01:24:11

see with these spherical harmonics, which

01:24:14

Right now they are compressing more, this is basically the RGB channel, so the more information, let's say that

01:24:26

When rotating the model, we can have more quality of reflections and brightness and so on, this just does not give it control, post-shot, luma generates it by default, Policam does not impregnate it and

01:24:39

He doesn't want to either and for now, but well this is going to change at some point, so chip away from here post-shot, well what

01:24:50

I recommend that you say if we are going to work on this side of post-shot directly, it is like a review of this, it is to decide if we want

01:25:01

use all the images or a portion, I would recommend that

01:25:06

we would use, because it is not the best, a maximum of 5,000

01:25:10

images or 50% depending on how many images there are, are values that

01:25:15

toss it in the air, but it has to

01:25:17

be like a part, it is better not to use all of them because

01:25:20

It would be too much, the other part is, well, deciding if they want to use them in maximum resolution, this also suggests that

01:25:30

Your video card will suffer more and the RAM memory will consume too much as you have noticed, so maybe, leave it at an average if your photographs are 4K, maybe put them in

01:25:44

2K, maybe half or take some number depends on what

01:25:48

want to do, if this is going to be a movie I would definitely leave it to the maximum, if this is going to be a web I would possibly stay

01:25:57

with the suggestion that

01:25:58

It makes me post-shot for the parameters that

01:26:00

has, I wouldn't touch this one here for the moment but I would investigate that

01:26:05

treats, as always, this value on this side what

01:26:12

does is that

01:26:13

generates or completes us the sky that

01:26:16

It is what

01:26:16

William spoke of that

01:26:17

It takes the profile of the last pixel of the horizon and completes the sky in a false sphere, made in 3D but within the gauchan splatting, so here you can prove that

01:26:30

It is more useful to them, if they keep the sphere or leave it finally if they put it, then they can remove it or if they do not put it, then it would not be complete with 3 of view, this value does not do

01:26:43

It needs to be explored but it seems that

01:26:45

It is very interesting here at the time that

01:26:48

3 of how to clean the models from burrs or clouds in post-shot

01:26:53

They are on the banks, this would be interesting to explore because

01:26:56

It is practically a mask that

01:26:58

generates you based on what

01:27:00

processed post-shot and so finally the sharpness part that

01:27:06

For me it is something very light but 3 of sight can hold more, so right now I go around 400 depending on the amount of images that

01:27:18

let's put in, here I put 400 images and 400 worked very well for me but this is the sharpness that

01:27:26

you could have in that final model and the church is that

01:27:29

You shared it with us, you made it 400, the church, let me see if I have it here, I'll show you, how much is it built? Here it is, let's see if it gives me the record of the parameters, that

01:27:45

We have, and well, something that

01:27:47

You're going to see right now that

01:27:48

This is part of the last thing that

01:27:51

I want to show you in relation to this, is that

01:27:53

Here in post-shot the model looks very clear, that is to say here I tried to clean it with the necessary caulk, but in 3rd view when I take it there this looks terrible, I don't know what

01:28:07

has to

01:28:08

See, when I put in this full model 3 of view like that

01:28:12

averages everything or has a maximum offset to display such that

01:28:19

It takes away quality, even, no, if I stay

01:28:22

only with the small church

01:28:24

controllable leaves me much more niggida, I don't know what it is due to, this is done, I don't know if it gives me the parameters here, not above I think that

01:28:40

you have to

01:28:40

select above in the, let's see if it is here are 450 k steps, 3 thousand by default you had put k steps of k spots, I left that by default and here the photos

01:28:53

Also, I didn't put them reduced and I didn't leave them in their total, this took 22 hours, more or less, okay, well, I'll take this one down in the training after, okay, this part here is not

01:29:08

the k steps, I had limited her so much, I mean I think that

01:29:12

From 30 by default one day I did a test 60, but not much more, up to 450 for this level of detail I am interested in practicing it, yes, but they have also here the sky, it is, I left it incomplete, no,

01:29:29

No, do not activate the option that

01:29:31

generate the sphere, when I activated it, in this case it generated an error, yes, in the same, that is, for the topic of the processes in pocho, if I already, I have done everything in pocho,

01:29:42

the last example, I already did the whole last practice, all in pocho because

01:29:46

I was giving it to the realitical components, so if I open the project now, which

01:29:53

I have done it with a 30k steps configuration, the same project has already aligned the cameras, source material and I want to retrain it, is it possible, or do I have to

01:30:04

redo it from scratch

01:30:06

It's something that

01:30:07

I need to try, just for example here, if I prescribe this part here, that

01:30:13

It's the Radiance Field, I don't know what

01:30:17

I'm leaving

01:30:18

I haven't done it because

01:30:19

Well, I've invested a lot of time in this moment

01:30:22

But if I put recipe here, I'm going to put it that

01:30:28

Yes, I'll see if it starts from the image, or only from what

01:30:34

I would be

01:30:36

There is something down there, if you put something to align in the amaras, we are already allied

01:30:40

The thing is

01:30:40

we started from scratch, of course

01:30:43

Yes, just what

01:30:45

It happens, right? So, and compose is very powerful

01:30:49

I say this because

01:30:50

yes that

01:30:51

we have seen that

01:30:52

potso suddenly lets you pause the workout and resume it later, that does let you

01:31:00

Yes, yes

01:31:01

And without resetting the Radiance Field, have you tried changing the 7, or is it not possible?

01:31:07

It's like thinking that

01:31:11

If we upload it, continue, return it again, I just returned it

01:31:16

Yes, what

01:31:17

suggests it is honest that

01:31:19

I put 500 here, let's see if it continues to advance

01:31:25

But there is like a button that

01:31:26

Say, apply something, right? Or maybe, who is there? Let's see, let's see

01:31:31

He tells me that

01:31:33

I lost a video, didn't I? If they put what they

01:31:35

have this computer, but it's a possibility, isn't it? Start exploring with things, are the programs that are simple?

01:31:41

Well, okay? Yes.

01:31:42

Surely the user experience to find

01:31:46

But for that reason, as I say so simply, it seems to me, there is no proof as to when you break the orders, to see if he reacts or you have to

01:31:54

start over from scratch because

01:31:55

it would be very good

01:31:56

If I have now done those 20 typical things, now I also

01:31:59

I had the computer for a decent result, to be able to say now, hey, I'm going to see the same project, to see how I can improve it to 400, from 60 to 400, but not having to

01:32:12

realign the image

01:32:16

Now, thinking about the workflow, so

01:32:18

understand what

01:32:18

We will have you, the ideal would be to start with something very light to see how it looks, because

01:32:24

If you pass, navigate and then go for the heavier option, right? You take it off with an object, instead of a building.

01:32:32

Yes, or with the same building in the dataset, upload it in what

01:32:37

It's better not to go to Madrid, there they scan, upload the dataset and in what

01:32:42

They arrive at their city, well, they already have it there, or the best one is there, this is a preliminary result, made in Kiri, in Polica or Luma, so that

01:32:51

can go exploring with what

01:32:53

They can be found before entering PostShot, right? The Realiti Capture option, I mean, is a great option, right now you would recommend them, if they don't want to get too involved, when they have some simple orbits, or

01:33:08

be some orbits taken in your project, let's say that

01:33:11

taken at once, from here I think that

01:33:14

it is very unlikely that

01:33:16

have any error and that

01:33:18

That, in truth, will provide you with an advance in both time and quality of your final project, but if you do it in several parts, well, what do you know?

01:33:30

can they face each other? I take what they do as homework

01:33:34

I said, about preparing the material for Realiti Capture, let's see if since I have several phases and several shots, and several techniques, I'm going to see if by naming the scans in order, we can make it easy for him and

01:33:52

Realiti Capture interprets it better

01:33:56

I'll see

01:33:58

Mario, can I ask you a question? Yes, go ahead, of course.

01:34:02

When we see this development of the transplata, we have not realized how

01:34:06

advances depending on the areas of interest for development, because

01:34:09

For example, at the beginning there was a lot of talk about the possibility of scanning self-transparent and self-transparent films, as if

01:34:15

came in as if to solve part of that, but I am also ignoring the Indo Paper that

01:34:20

They talk about the problems, I don't know, unchained, intermittent, image optimization, the issue of recording movements, the blur issue, but I have a specific problem that

01:34:29

I think that

01:34:29

All of us are concerned or interested, it is as you see the issue of interior scanning, that is, not orbiting an object but

01:34:39

upside down like a reverse orbit, as if you were standing still, recording between 3D upside down from you, to be able to, for example, capture one room then another and use them as models to then do your rest of the work

01:34:52

and do interviews in Unreal Engine or whatever

01:34:54

be, because

01:34:55

For now, of course, capture any object, no matter how small

01:34:57

or an object that

01:34:59

Orbiting it in a drone is pretty much a done deal, but how do you see the development for scanning and beyond? Yes, I shared with you the previous time, a case of a taco shop.

01:35:12

Made by Google, it seems to me like PinCalifornia, it's a small restaurant

01:35:18

where the camera came in and out, it was practically a cameraman who

01:35:22

I went in and out of the taco shop

01:35:24

and I already had the whole interior stage complete, the technology that

01:35:29

They use Nerf, it's this other one, but I think that

01:35:33

Yes, you can walk there, they use this technology, it is a very simple process and you get a raised space in coming and going, practically, if we omit that

01:35:47

Google processes with a large number of computers, that will not be an issue for them but practically the one that

01:35:53

Someone comes in with a backpack and explores the corners and comes out, not even talking about this type of device, something much simpler like a 360 camera, right? The promise of 3dista that

01:36:07

It hit us in Granada, I don't know how long ago, well at least that, that is, they told us that

01:36:13

had a 360 camera and you can generate models, I think that

01:36:16

It happened to all of us, we invented at that moment that

01:36:18

I had the 360 camera and you understood the model of this dictation and so far I see that

01:36:24

is what

01:36:24

is even further away

01:36:26

But Luma what

01:36:28

I said, Luma lets you upload 360 photos and I don't know if video, if it doesn't let you video, you upload a sequence of frames that

01:36:35

irregular, the Luma

01:36:37

And how are you, Luma? Look, an interior there? In this video, Luma

01:36:42

What? I'm lost.

01:36:44

Luma doesn't let you upload 360 video.

01:36:46

That's why I say that

01:36:47

which

01:36:48

León said that 3dist is getting Luma for now

01:36:52

Yes, in fact I don't know if you knew that

01:36:54

Luma has a solution just for interiors, it's the same, these are splatting, but with a video

01:37:00

But Mario, I was looking at this just last weekend and it seems like something that

01:37:06

we need it, because

01:37:08

This is what

01:37:09

It does a splatting and then generates an automated video, but it doesn't let you get involved, however if you go to Luma, it generates the Gaussian splatting and then you animate the camera

01:37:21

as you want

01:37:22

And that's why he's charging you.

01:37:24

Correct

01:37:25

Yes, but I have it here.

01:37:27

No, no, look

01:37:27

Yes, not at the moment, at the moment it is correct

01:37:31

Here is a spatially captured video that

01:37:34

then you can do reframing or sharpen the shots or make them straighter or something

01:37:40

It doesn't seem like it to me

01:37:41

This part has been more commercial, more than

01:37:43

work for us

01:37:46

I think that

01:37:46

it is more commercial

01:37:47

Hence

01:37:48

I said, I gave this a test and I said, but if I want it in HD they're going to charge me and such, I saw it as more commercial, more for the end user

01:37:58

Final

01:37:59

Yes, but well walked around here

01:38:00

So, just like that

01:38:01

I think that

01:38:02

This technology that

01:38:02

use Google, which

01:38:03

Here I proposed Flytrux from Luma, it is aligned with that new thing, but I think that

01:38:10

3 of view still in this software that

01:38:12

is going to launch in the next 2 months, well yes I promised something like that, that

01:38:16

with a 360 camera, which

01:38:18

Even some of the moderators said, we're even going to tell them that

01:38:23

camera we use or which one we are going to recommend for you

01:38:25

do that

01:38:27

I actually give you the spoiler here that

01:38:30

I just asked them today, that

01:38:33

Jose on Facebook has said several times 3 of vista extens, 3 of vista extens and I say, they are naming it as that

01:38:43

outside the family and they haven't taken it out

01:38:46

So I say to him, hey, have you mentioned the extens thing several times? There's something, something that

01:38:50

Can you tell us what

01:38:51

the story goes, because

01:38:53

We are eating our heads here that

01:38:55

I continue with credits, with cloud, with processing, something local, we have no idea free or paid

01:39:02

And then I told him, hey, with Mario, we met with José almost a year ago

01:39:08

We could get back in touch with José and that

01:39:11

Let us know here in privacy how things are going, right? This kind of thing that

01:39:17

we are talking

01:39:18

That part is amazing too, isn't it? Let's see where it goes, but I think that

01:39:22

Yes, it's all a matter of time

01:39:25

Just look at that

01:39:26

It says beta here, so yes, that is something that

01:39:30

not ripe yet, right? Yeah, here's something that

01:39:35

What worries me at the moment, 3rd sight, is that

01:39:36

Look, here I have these elements fairly clear, this model as is, I import it to 3rd view in this exercise that

01:39:47

is over here

01:39:48

Yes, and look, it's exactly the same model.

01:39:51

I don't know what's going on

01:39:52

That 3rd view compresses it

01:39:55

I mean, I don't know if it prorates the total of the scene and limits it to not being a million splats

01:40:00

I would report, Mario, to see what

01:40:03

tell you specifically with your project

01:40:06

I would send you a report

01:40:07

With that whole image, right? Yes, I would send this project to them and you admire it because

01:40:12

suddenly they give you an update and it will reveal to you

01:40:16

Or not, or not

01:40:17

I mean, no

01:40:18

I think that

01:40:20

They are already compressing a lot, and that is the reason why

01:40:22

You cannot upload compressed files, because

01:40:24

They can't re-compress, I think.

01:40:26

has to

01:40:26

see

01:40:26

Did I do this with the Mario-level beta or something? This one's with the new one, with the latest Java.

01:40:31

What did I do? I opened the project with the beta because

01:40:34

the beta adds an update

01:40:37

Now there was a time when

01:40:38

The official version was ahead, but now it's back in the latest updates and has entered the beta.

01:40:45

Try making a copy of the project and opening it in beta to see if it does the same thing when exporting.

01:40:50

Yes, here, for example, what

01:40:52

what I did is that

01:40:52

This church cuts it, exports it as is, that is, without doing anything to it

01:40:57

And in this case, I think that

01:41:00

I'm going to change it here

01:41:03

I think that

01:41:03

Yes, it improved a lot, because

01:41:05

we have a smaller model

01:41:06

Well, this is the 3D model of the Poli, but the small one

01:41:09

this would be more good

01:41:11

This is a small Gausham,

01:41:12

So here, there was no compression at all, right? I did it by the total number of Splats

01:41:19

I believe

01:41:20

And that's it

01:41:20

It's kind of limited, isn't it? The extension has more, it makes more Splats, of course.

01:41:25

That's right, here I have a medium version, so

01:41:28

I erased everything necessary on the horizon, the mountains and part of that and it also behaves much better

01:41:37

But there you see that

01:41:39

It's still preloading, right? It depends.

01:41:43

It is a dynamic resolution

01:41:44

I want to think

01:41:45

But yeah, I didn't like that very much.

01:41:48

the complete model, compress it so much that

01:41:51

I understand that

01:41:52

This online is going to see everything that

01:41:54

is recharging

01:41:55

And we've been going for like five seconds, ten seconds at a time.

01:41:59

It doesn't work, right? Maybe I don't want to have it online, but locally, but

01:42:04

this could be seen or what

01:42:07

This is like part of this scheme,

01:42:12

No? So, well, I don't know, up to here, if since

01:42:17

It took us quite a while, have some other comments, a question or something specific

01:42:35

I don't know if I don't hear it

01:42:37

Look, I do, I show this quickly, a yes, of course

01:42:41

And how was I scanning in this

01:42:48

I'm going to stop sharing

01:42:51

This one along with Oscar, at the Climbing Wall, this that

01:42:58

spoke

01:43:01

The order an interior

01:43:03

If you are standing here at one point and you are turning

01:43:11

No, no, no, it was because

01:43:12

I was going to take the position and it was to advance

01:43:14

There's Oscar, down there.

01:43:15

So, the thing was to move forward

01:43:19

To scan diagonally down and diagonally to be parallel to the

01:43:27

The trajectory is parallel to the wall, but the drone is also turned to the left.

01:43:33

And I did that for there and back.

01:43:36

But then your goal was

01:43:39

sorry, scan it, just the wall

01:43:41

Of course, the environment

01:43:43

There again

01:43:46

It's still like an object in the background and it's orbiting that wall

01:43:50

Sure, sure

01:43:56

Of

01:43:53

And Guiel, haven't you made the model yet? No, no, I don't have time for this one yet.

01:44:02

I also made this wall as

01:44:06

I would go to a corner and

01:44:09

You see, I took the center and turned the drone

01:44:13

You've seen it there, haven't you? I'm orbiting to catch the wall from different angles.

01:44:21

There is for example also

01:44:27

Of course, what

01:44:29

I don't know how people are moving, how they will end up

01:44:32

in the Gaussian, but well, they will surely remain

01:44:34

medium borrones

01:44:36

But with the interior model, it goes to 360, Guillermo has made us then

01:44:49

How are you going to 360? What do you mean?

01:44:51

Was it achieved, for example, with a 360 camera? No, no, no

01:44:56

I haven't done the test, have I?

01:45:04

And then, if we wanted the building from the outside, also in several phases, and this that

01:45:14

It was at Oscar's request, I have processed this one.

01:45:17

We orbited the sign

01:45:25

Yeah, like that

01:45:25

Orbits always tend to work better when processing, I don't know, I did it at two angles, much lower, and then this one more of the building, this is a normal video to give it to, this time

01:45:46

from below to grab the roof, and I also decided to do a lot of vertical work, since

01:45:57

The drone allows me to scan, I think so.

01:46:01

I do it faster vertically, because

01:46:04

heals more area

01:46:06

Right? Yes, and what

01:46:16

I have seen that

01:46:17

I have also included vertical and horizontal formats, also in Boss Shot and I have taken it

01:46:23

Of course, then here I take the sky and I take the ground, and I don't even need the environment anymore.

01:46:30

It has this wall, here unscaled, so I was also interested, there was the tree, and that's why I then flew down here

01:46:47

Mario, can't you share the link to the presentation afterward? Yeah, sure, yeah, sure, it's online, the link was shared, already in the last meetings, right, Leon? It's the thread here, I'll do it.

01:47:06

What are you asking, Oscar, in the, in the first meeting at least, there was the shared mirror

01:47:15

But if you want to put it here in the chat, yes, of course, I'll share it.

01:47:20

Let's see if you can inform me if they can access

01:47:25

Let's see if they open it, maybe you got a notification from cogracio before, I was trying to get in and it's possible

01:47:34

Ah, what

01:47:36

Well, if you don't need to register or anything, it's enabled to view

01:47:40

In fact, it took me another day to get a notification that

01:47:43

you had put it by hand, that

01:47:44

you had made a modification

01:47:46

Ah, yes, he snitches, he's a snitch

01:47:50

Well, thank you very much, Mario, I think that

01:47:54

another good review and I think that

01:47:58

like everything, at least to me, now the part that

01:48:00

It's my turn, as in other areas, it hasn't touched me, what

01:48:03

I experience that

01:48:04

as I'm hitting it hard because

01:48:07

I am offering it, that's why I may have so many doubts

01:48:12

But I think that

01:48:12

I'm going to touch it, touch it, touch it, go out, do, what

01:48:16

I made Leon practice with objects, which

01:48:19

It's going to be easier than

01:48:20

Flying a drone over a church and layering it, once you're getting the hang of the tool, you can enlarge the object, right? Right, there's a theme that

01:48:36

We don't touch it, it's the editing part to be able to clean it up.

01:48:40

Enriquezco

01:48:40

also, yes

01:48:42

There what

01:48:42

I'm going to do more like, I owe Pietro a video for this, right? So I'm going to prepare it for everyone, very compact, what I'm going to do is

01:48:49

Right now I saw it's Super Splat again, it's back in play, so I definitely know that

01:48:55

It gives you more control to edit and two to mix more than one splat in post shot, that is not possible, that's where I got to in that part, I don't know if in the latest version or something, but that

01:49:10

It is possible, I think

01:49:12

It is possible, it already allows you to put several objects, I think that

01:49:16

I've done it with potso, yes, excellent, and one thing, a small one

01:49:21

counting that

01:49:22

If you use Kiri you can export it in Met, and there you will obviously be able to clean it up and work on it, and if you do it in Gauchan it also has a plugin for Blender, when you can also work on it in Blender,

01:49:33

As an alternative, I have obviously found that

01:49:35

Super Splat is a better option and all, but why

01:49:38

It is known that

01:49:38

It can also be done by Blender

01:49:40

Well Mario, I know that the video

01:49:42

says is that

01:49:43

You have Pietro watching, please send him to me so I can upload him to the resources.

01:49:48

I share it with you, I am going to take some time today if possible, and only from that activity that

01:49:55

It's the ability to clean up a Gauchan so you can save it, import it into Vista 3, or even merge it with another, right? Perfect, very good, time flies here, and the rest of us had a lot of husbands, what's up? Yeah, well,

01:50:15

It's time to eat here

01:50:17

Of course, it was before knowledge

01:50:19

And here I'm just missing my coffee today.

01:50:25

Thank you very much, Mario, it was very good.

01:50:29

Yes, always, always a pleasure, and I hope that

01:50:34

we came to understand half, because

01:50:37

There are many technical aspects, but at least we can keep ourselves updated on where this is going, capable of

01:50:43

Next week we'll forget everything, it's done with a click, it can happen at any time

01:50:53

I don't know, anyone else, I don't know if you have any questions, if not, well, we're always in communication in the group, right? Thank you for all the support and conviction that

01:51:04

Share it with us, Mario, thank you very much

01:51:07

Thanks Roberto

01:51:09

Roberto is a faithful follower lately

01:51:13

He's there in the shadows, gathering all the knowledge.

01:51:19

Excellent Horacio, a hug

01:51:23

Okay, guys, thank you all very much, and Mario in particular, give it a try, I'm also going to share, as I progress with the professional example and these other practices that

01:51:36

I can go and do them, I'll share them, because

01:51:39

I think that

01:51:40

with those doubts that

01:51:41

they are coming out, it is where everyone is, because

01:51:43

Suddenly we have Mario as a guru, but he was simply the one who

01:51:48

He has stung us here with the poison

01:51:50

And he is more like too

01:51:53

Let's all learn, right? There, there, I'll give you a sneak peek later, I'll also share an exercise where Marco helped us take some videos for Monterey and Guillermo helped us with the programming.

01:52:05

three of view or if you call alumina, I share a group with you so that

01:52:09

see how it turned out,

01:52:10

And all comments are welcome, right? Bravo, excellent, excellent, thank you.