Latest / Bad Decisions Podcast / #34 CEO of Polycam on Democratizing 3D Capture, 3D Gaussian Splats and AI | Bad Decisions Podcast
Transcript
- 0:00Chris Heinrich, CEO and Co Founder of Polycam, with a
- 0:04vision of democratizing 3D for everybody, everybody, everybody.
- 0:09Polycam is a 3D scanning application that allows you to
- 0:12capture the real world and turn it into 3.
- 0:14D Anyone opens Polycam today, they will see Lidar, they will
- 0:18see. Photogrammetry.
- 0:19They see rue. So how did you start, Polycam?
- 0:21I joined a startup called Ubiquity 6.
- 0:23That's where I met my Co founder Elliot.
- 0:25Our mission is really to democratize 3D creation.
- 0:28I started working on Polycam the day after.
- 0:30I got laid off couple years, we'll see some big improvements
- 0:32for for Gaussian splat. The research community, everyone
- 0:35took note of Gaussian splats during SIGGRAPH and then we
- 0:37started building on top of it and making these improvements to
- 0:40to remove floaters. Welcome back, everybody.
- 0:42Chris, it's good to have you on the podcast.
- 0:45How are you doing today? Good.
- 0:46It's great to be here. Thanks for inviting me on.
- 0:49Really excited to talk about. Computer Graphics Polygam.
- 0:52We have to say this, Chris. We've been meaning to speak to
- 0:56you for the longest time and we have to give a shout out to the
- 0:59one and only barriers. Shout out to barriers or making
- 1:01this happen when we when we met Barris online through Polycam's
- 1:06channel of course, a couple of months ago when we were using
- 1:08Polycam for our photogrammetry for the scenes in Unreal Engine.
- 1:11Five was Meta Human that time was Meta humans as well.
- 1:14We got connected with Barris and we told him we want to talk to
- 1:17Chris and he told us you have to understand, Chris is very busy.
- 1:20He's working on a lot, a lot of stuff.
- 1:22And you know, getting him on the podcast is not going to be that
- 1:24easy. So we said, OK, you know what,
- 1:26we'll wait, we'll wait. We'll wait.
- 1:27And we waited. And I have to say, Farad, we
- 1:30have Chris on the podcast and I just want to say thank you
- 1:33because we understand you're a busy man.
- 1:35You guys are doing so much with Polycam.
- 1:36Plus we know this. You guys dropped the 3D Gaussian
- 1:40splatting technology. And the new update, which for
- 1:43anybody that doesn't know it has less floaters, which we're going
- 1:47to discuss in a moment, but we just want to say thank you for
- 1:50giving us the time today. Yeah, no, it's it's my pleasure.
- 1:53I mean, I love talking with people in the industry, people
- 1:55that are pushing the edge of technology.
- 1:57I'll have to thank Barris for being my, being my manager in
- 2:00this situation, playing, playing hard to get for me.
- 2:04But no, honestly, I'm. I'm excited to chat with you
- 2:06guys. Thank you.
- 2:07So let's start with something that is a huge mystery for I
- 2:11believe not just me and Farrah, but almost everybody who uses
- 2:14Polycam. We want to know about the origin
- 2:17story of Polycam. And before we even get to that,
- 2:19and before I let you even answer that, I want to tell you my
- 2:21origin story with Polycam. This was a couple of years ago,
- 2:25just I think actually just a few years ago in Malaysia when we
- 2:28were working on a 3D project within Blender at the time.
- 2:32Not unreal yet. We were just looking for ways to
- 2:34increase the efficiency of our workflow and also have more
- 2:37photorealism. And then I believe it was a
- 2:39friend or I saw it online where I heard about Polycam and they
- 2:42told us it's so easy, just get an application, you know, take a
- 2:45couple of photos and you have them on.
- 2:46I was like, Nah, no way. It takes me so much time to to
- 2:48model this. And at the time, by the way, I
- 2:50didn't know anything about photogrammetry.
- 2:52And so me and Farad, we got Polycam up and then it was super
- 2:56easy. Exactly what people said.
- 2:57He was intuitive and within minutes we got the model I I
- 3:01believe it was Tangiro from. I had like an action figure
- 3:06model and I just took a couple of photos and my razor mouse and
- 3:10I think it's in my profile as well.
- 3:11Maybe I'll pull it up later for you guys to see.
- 3:13And then we got that in Blender in a couple of minutes and it
- 3:17looks so good. And that moment, the first time
- 3:20I think people do photogrammetry is such a magical moment,
- 3:22especially if you're doing 3D, because it you just see how much
- 3:25time you can save and how many cool things you can create.
- 3:29And now of course the the use cases are being explored by
- 3:33various other members in the community and it's crazy to see
- 3:35all that stuff. But I want to go all the way
- 3:37back to your story. How did you get into all of
- 3:41this? Why Polycam?
- 3:42Why photogrammetry? And why all of this actually
- 3:45happened in your life? Right.
- 3:49Yeah. No, I'm.
- 3:50It's really interesting that you say that because it reveals this
- 3:53kind of like expectation that people have that like doing
- 3:55stuff in 3D is going to be hard. Like, you know, you can take a
- 3:58photo and you can pull that into a dozen apps on your iPhone and
- 4:01post process it. You apply AIFX and it's really
- 4:04easy. And people aren't impressed
- 4:05anymore because we expect that to be easy.
- 4:07But with 3D, there's this expectation that it's going to
- 4:10be hard. And that's kind of like what
- 4:11what Polycam is all about is trying to democratize it, trying
- 4:13to make it, trying to make it easier, and trying to make it so
- 4:16this next kind of generation of creators.
- 4:19I mean, eventually it'll be taken for granted, but if you've
- 4:21done your job well, then the ease of use is something that
- 4:24will be, you know, taken for granted.
- 4:27So that's, I'm glad to hear that's how it played out with
- 4:29you guys and hopefully, you know, it plays out over the next
- 4:325 to 10 years with with lots of other people.
- 4:36But my interest in 3D scanning actually dates back to like 2016
- 4:39or 2017. So I was, I was finishing my PhD
- 4:44in physics. I was kind of.
- 4:47There's not. I was kind of looking out for
- 4:49for what's next for me. I I I've been in physics
- 4:52academia long enough to know that I didn't want to stay in
- 4:54physics academia. The job market's not that.
- 4:56Not the the The job market's not that great.
- 5:02You kind of have to wait for a tenured professor to die to get
- 5:04a job. Job market doesn't increase.
- 5:08So I knew I wanted to go somewhere else.
- 5:09I mean there's. And there's a ton of stuff
- 5:12happening in technology and I think if you remember back in
- 5:142017, there was actually a big hype cycle around AR and VR.
- 5:18The Oculus was coming out, the HTC like the first wave of of of
- 5:22VR headsets was coming out. Also, Pokémon Go.
- 5:26Oh yeah, Oh, the zombie apocalypse.
- 5:29Everybody going out to like multiple phones catching them.
- 5:32Yeah, yeah, yeah. Exactly like Tim Cook's out
- 5:34there evangelizing AR. It felt like the 3D future was
- 5:37just around the corner, like 29. Everyone was like Apple's going
- 5:40to come out with AR glasses in 2019 because Google Glass had
- 5:43come out in 2014 or whatever. And I was, I was at the end of
- 5:47my PhD and I was kind of had my eyes.
- 5:49You know I was, I was looking out for a problem that I thought
- 5:51would be interesting and valuable to solve because I knew
- 5:53I wanted to go into technology entrepreneurship and it seemed
- 5:56like 3D scanning was just this super valuable problem because
- 6:00if we're if we have this 3D future.
- 6:02With these 3D headsets then we're going to need 3D content.
- 6:05Like where is all that 3D content going to come from?
- 6:07Like I knew enough to know that it was.
- 6:09It was made sort of manually and be a very slow process in tools
- 6:12like Blender and Maya. And it seemed obvious that, you
- 6:16know, if like look at on the Internet, almost all the 2D
- 6:19image and video content is captured from the phone.
- 6:22And it seemed obvious that the same thing was going to be true
- 6:26with 3D, even though. Back back then, like
- 6:29photogrammetry wasn't that good. There were these, like people
- 6:31were still using these like $40,000 laser scanners.
- 6:35And so I got, I got interested in the problem at that time, and
- 6:38it was actually it it took longer.
- 6:40Like it often does, kind of, if you're starting something a lot
- 6:42longer than you expect to actually get to a point where
- 6:45that that sort of dream and that vision starting to be realized.
- 6:49So in my case, I started a company after grad school that
- 6:52was originally intended to be a 3D scanning company.
- 6:55We were going to 3D scan products for e-commerce
- 6:58companies that were going to power their like VR shopping
- 7:00sites or VRAR shopping sites. And it it turned out that for a
- 7:06variety of reasons, it was just more economical to actually have
- 7:093D modelers create their 3D models of their product catalog,
- 7:12and we can just send product photography to 3D modeling
- 7:15studios. And they can churn out really
- 7:17good models. And then we ended up building
- 7:19just AR apps that kind of delivered that content.
- 7:22So like AR shopping apps for a variety of companies.
- 7:24But it was, it was 2018, it was, it was there was a lot of
- 7:27interest in this, but it was, it was too early, you know big we
- 7:32sold a couple of like innovative e-commerce companies, but most
- 7:35of the big e-commerce companies were like that's interesting but
- 7:38we are going to, we're going to wait.
- 7:39And I mean they made, they made the right call and I kind of
- 7:42fell for the hype in some sense. That was kind of a lesson
- 7:44learned. My first foray into technology
- 7:48was like getting really excited about this hype but not focusing
- 7:51enough on like what applications are ready for for prime time.
- 7:54And then when we get back into like nerfs and gaussians
- 7:57splatting, that sort of experience, it sort of shaped me
- 8:00now to look at like new technology with a more critical
- 8:03eye to make sure that it's, you know, we think that there's
- 8:06real, real value to be unlocked. Fast forwarding.
- 8:10I joined a startup called Ubiquity 6, which.
- 8:14Is really where Polycam's story began because that's where I met
- 8:17my Co founder Elliot, which you'll have to have on the
- 8:19podcast sometime as well as as well as two of our first
- 8:23engineers that we kind of started the company with and
- 8:26Ubiquity 6 originally was also an AR company.
- 8:30They were, they were an AR Cloud company, which people don't talk
- 8:35much about the AR cloud now, but in 2018 like the AR cloud was,
- 8:38was the thing that people thought.
- 8:40What do you mean AR Cloud? The AR cloud, right, is this
- 8:44idea that, like, we'll all have AR glasses and that there's
- 8:46going to be this digital layer that's kind of on top of the
- 8:49real world. Yeah.
- 8:51And you can put your glasses, you can walk down, you can walk
- 8:53down the street and you'll be able to see like digital
- 8:57graffiti that your friend or somebody else, or like the menu
- 9:00of the restaurant right in front of the restaurant.
- 9:03Prototypes like you look at one product and you see the
- 9:05ingredients, like your allergies and everything will pop right in
- 9:08front of you. Yeah.
- 9:10It was like or like the AR Cloud or AR Internet, The idea that
- 9:13you're not just looking at this Pokémon that's like only you can
- 9:15see. But there's this kind of
- 9:16persistent digital layer that's over, that's laid over top of
- 9:19the real world. And there's this kind of they
- 9:21call it the AR Cloud because that all that content actually
- 9:24lives in the cloud, but it's kind of realized and viewed in
- 9:28the real world and what what they realized when they wanted
- 9:33to. We're embarking to kind of
- 9:34create the AR cloud is is realize that in order to hang AR
- 9:38content on on the real world you actually need a a 3D map of of
- 9:43the real world to align that digital content to.
- 9:46Same thing. Like if you're in a you know if
- 9:48you're in you know 3 DVFX software right.
- 9:52You have to have your rough 3D model that you're hanging your
- 9:54kind of your your 3D content against your virtual camera
- 9:57attract the object so that you can place it on top of it.
- 10:01Exactly. And so they were like Oh no, we
- 10:03we need, we're going to need all these 3D maps of the real world.
- 10:05How are we going to do that? Let's let's release the 3D
- 10:07scanning app and then Crowdsource 3D maps of the real
- 10:10world so that way then we can put AR content on it that was
- 10:14they kind of backed into 3D scanning from that point of
- 10:17view. And but ended up releasing an
- 10:21app called Displayland, which was a really great, you know,
- 10:24early 3D scanning mobile app. It was like kind of the first
- 10:26one that had mass market appeal and even though it was like
- 10:29faster and. Not quite as easy to use as
- 10:32Polycam today. I think for a lot of people that
- 10:34that used it, it had that similar feeling of like Oh my
- 10:37gosh, photo photogrammetry's a lot easier.
- 10:39I'm just taking a video on my phone, uploading it to clouds,
- 10:41taking care of it and then I get my 3D model back.
- 10:44And that was you know you know some people are are fortunate
- 10:49enough in my career to have that experience going into Polycam
- 10:51and be able to kind of take that experience doing something very
- 10:55similar for for a similar market.
- 10:57And then apply it kind of in my own way to to a new market.
- 11:00And that's kind of what happened with Polycam.
- 11:03They they Ubiquity 6 didn't manage to find quite the sort of
- 11:08market fit that they had hoped for as a company.
- 11:11And during the pandemic kind of like I think a lot of companies
- 11:14that were not, they were unsure of their position in the market.
- 11:17They kind of pivoted and changed direction, laid off half the
- 11:21company, including me and my Co founder.
- 11:23They did something unrelated to 3D, they did something related
- 11:26to video and then ended up getting acquired by Discord.
- 11:28So it was a good move on their part, a good happy ending for
- 11:32you know the the half of the company that that got to take
- 11:34part of that. But for me, I got I got laid off
- 11:38along with like the rest of the 3D team, kind of in the middle
- 11:41of the Pandemic. And I started working on Polycam
- 11:44the day after I got laid off basically.
- 11:47That's that's fascinating. The dots have been.
- 11:50Connected I. Know I searched all over
- 11:52Internet, I couldn't find it. I couldn't find anything on this
- 11:55topic. Yeah, this is this is crazy.
- 11:57I have so many questions. But Farrad, I know you haven't
- 11:59too. So what you just I'll.
- 12:01Just ask one question. Because you started Polycam,
- 12:03what was the first product that you guys go with?
- 12:05Because if anyone opens Polycam today, they will see Lidar, they
- 12:09will see photogrammetry, they see room.
- 12:11Yeah, 360, you'll see so many things.
- 12:12What was the first thing that you guys worked on?
- 12:16Yeah, it was. It was LIDAR scanning.
- 12:17That was the thing that put us on the map.
- 12:19And we just got really lucky with the timing there because
- 12:23the Apple added this LIDAR sensor to the iPad in March of
- 12:272020 and I got laid off in June of 2020.
- 12:31And there were rumors that they were going to bring it on to the
- 12:33iPhone in, you know, September, October of 2020, you know,
- 12:36during the normal iPhone cycle. And I had actually been working
- 12:39with the LIDAR sensor on the iPad internally at Ubiquity 6
- 12:42before I got laid off. I was even advocating to the
- 12:44leadership there that hey. You know, this is a big deal.
- 12:46The LIDAR sensor is going to open up the market a lot.
- 12:48They had already kind of I think picked, picked their path.
- 12:51But it it meant that by the time that I got laid off, I had, I
- 12:54was thinking about, you know, starting a company focusing on
- 12:58the LIDAR sensor even before I got laid off because I knew it's
- 13:02not like, you know, Google had added a sort of a depth sensor
- 13:06to some of their phones. Like there was this Project
- 13:07Tango, which maybe, maybe you've heard of, but.
- 13:12Yeah, but when when Apple puts some a sensor on their devices,
- 13:16you know it's part of their like 5-10 year strategy, right.
- 13:19So that was this clear signal that we're going to have this
- 13:21new sensor to work with. That'll be a big deal.
- 13:25And if I if it had been up to me, I would have waited until
- 13:29maybe October to leave the company and start my own
- 13:31company. But that decision was made for
- 13:32me, which turned out to be turned out to be perfect.
- 13:35In retrospect, that was. That was the perfect timing
- 13:38because I had three months to build sort of the first version
- 13:41of of Polycam and get it out on the App Store, just like a week,
- 13:47a couple weeks before the iPhone came out with the LIDAR sensor.
- 13:50We were the first app that used the LIDAR sensor for 3D scanning
- 13:53and it was kind of like instant, instant business.
- 13:57There was, yeah, there was pretty instant, like excitement
- 14:00and value. From that, from that first
- 14:04version of the product, how did you feel when you got laid off?
- 14:07And I mean, you had the idea, of course, of starting your own,
- 14:10but this happened before you were ready.
- 14:13I know that, you know, a lot of people might go through the same
- 14:15thing. I just want to get that message
- 14:17from you. How you dealt with that?
- 14:19Were you extremely motivated the next day?
- 14:22You were like, you know what, I already wanted to do this
- 14:24anyway. Might as well do it earlier or
- 14:26was it tough? How?
- 14:27How was that moment for you? Yeah.
- 14:31So for me, it was a little bit extra challenging because I was
- 14:33actually in the middle of buying my house, this house, and well,
- 14:38you're in it now, so you know. It worked.
- 14:42It was a happy ending. But the way the way that,
- 14:45particularly during the pandemic, the way that buying a
- 14:46house works, right, is you put down your your escrow money, you
- 14:50know, whatever your deposit of like $30,000 and then you then
- 14:53you have to close in three weeks and then the bankers are
- 14:55calling. Your they only call your
- 14:57employer like the day before close and say hey is this guy
- 14:59still employed right. And so I was.
- 15:02So for me it was extra stressful because it was like oh shoot,
- 15:05I'm going to lose $30,000 and lose this house.
- 15:08Fortunately ubiquity 6. I I kind of pleaded with the
- 15:10management team there to like nominally extend because I was
- 15:13originally the closing date was going to be like.
- 15:16Two days after my last day at Ubiquity 6.
- 15:26Yeah, it was so, so I was so that first kind of, you know,
- 15:30day or so, I was just freaking out about that.
- 15:32But fortunately the management team, you know, they extended my
- 15:34employment in extra 2 weeks and they agreed to say like, hey,
- 15:36when the bankers called that, hey, you know, Chris is still
- 15:38working here, which I was, but I think you know that, that aside.
- 15:45I had. I mean I don't know I think when
- 15:47you get to a, you know I'm, I'm 34 now so you get to a point in
- 15:51your career you've had ups and downs, right.
- 15:53I think those first downs they can really throw you down and
- 15:56they can be discouragement, be a discouragement.
- 15:58I think this time I sort of suspended my judgment a little
- 16:02bit because I I was like I don't know how this is going to work
- 16:04out in the end right. You you you never do.
- 16:07You have something that looks like it's bad and you can you
- 16:09can turn that into an opportunity or you can.
- 16:13Or you can get down about it in which case it it has a bigger
- 16:15negative effect. And I think just being a little
- 16:20bit later you know a little more mature in my career I was able
- 16:24to see that hey like this could this could be something really
- 16:26good and then kind of went went all in on that and and then it
- 16:29turned out really well. And I think that's a lesson that
- 16:31applies for you know, most anybody that's going has some
- 16:33unexpected downturn of events to like look you know, look, look,
- 16:39look for the upsides maybe it's, maybe it's.
- 16:42The universe, you know, telling you that like you should, you
- 16:44should be doing something else and and look look for that
- 16:48rather than just, you know, focusing on on what you what
- 16:50you've lost. Dude.
- 16:51That is literally the rule that we live by.
- 16:54Everything good comes out of something bad.
- 16:56Everything good. That's why we are called Bad
- 16:58decision studio. That's literally why every if if
- 17:02you see people losing weight, it's usually because of a
- 17:05breakup or someone made fun of them, a bad moment that leads to
- 17:07them losing weight to get fit, to go to the gym.
- 17:10If you see someone starting a new businesses because they
- 17:13they're in a bad situation, you know, all of these things sort
- 17:16of originate where something bad happens to you.
- 17:18But as long as you don't give up, right?
- 17:20As long as you suspend the judgement, I love that word.
- 17:22You just don't don't judge yourself too quickly And then
- 17:26you you just keep going and good things will come out of it.
- 17:28I love that. Thank you so much for sharing
- 17:29that. Yeah, for sure.
- 17:33And so then, I mean the rest of Polycan was sort of sort of
- 17:37history from there we added. LIDAR and pretty much all you
- 17:40could do is you could make a scan in LIDAR and then you could
- 17:42export it. I think we had a couple formats
- 17:45in the beginning and then most people would continue their
- 17:49their 3D creation flow and whatever tool that that that
- 17:51they used. If they were architects, maybe
- 17:53as a tool like Revit, if they're, you know, creators,
- 17:55it's probably like Blender and then over time, you know, we
- 17:59have added. More capture modes.
- 18:01As you've mentioned, we added a cloud based photogrammetry mode.
- 18:04We added three 6360 camera, which I feel like is an
- 18:08underappreciated feature, but we are like the only iPhone app
- 18:11that lets you get a full 360 on on the App Store.
- 18:16We do that using the Ultra wide camera to get you all the way
- 18:19around and then you're left with the top and the bottom of the
- 18:21sphere that we fill in with stable diffusion.
- 18:24And we added also room mode, which is a more sort of
- 18:27architectural feature. That automatically generates a
- 18:31CAD model and. Didn't we use room mode that we
- 18:34were renting the house in Canada before, So we used, so the house
- 18:38was empty and we went inside the house.
- 18:40We used the room mode, Pharos took out the office part, and
- 18:43then we started buying computer tables and desks and we said,
- 18:47OK, can we fit that in? Oh yeah, I think we can fit that
- 18:49in. Yes.
- 18:49And and then we did something extra.
- 18:51We at the top of that we also scanned that same room, brought
- 18:56it to Blender and then added the lighting before we went moved
- 18:59into the house to see what sort of lighting we should get for
- 19:02the place. So all this sort of cool shit?
- 19:03Through the artists, rent a house, by the way.
- 19:05Yeah, all all this sort of cool shit that you can do.
- 19:07And we've seen this, by the way, online.
- 19:09That's how we did it. We didn't come up with these
- 19:11ideas. They're all over Instagram.
- 19:12Sometimes you guys share them on your page as well of people
- 19:14using Polycam in a really cool way.
- 19:17What I want to talk about is actually the vision as well.
- 19:20I mean there's a lot of questions I have about the the
- 19:23speech that we just had in over the past few minutes.
- 19:26But let's talk about the vision for Polycam.
- 19:28You just mentioned we have the LIDAR, we have photo, we have
- 19:32room, we have 360 now 3D Gaussian splats.
- 19:36What is the ultimate vision for you?
- 19:41Right. I think you if you had asked me
- 19:42that like a year ago, I probably would have said something like,
- 19:44you know, 3D capture for everyone making.
- 19:48I think that in fact that's what it still says on our website.
- 19:50It's making 3D capture easier. And 3D Capture is still this
- 19:53like really important part of our product.
- 19:55You you launch into 3D Capture. I think we're very much known
- 19:57for 3D Capture, but over the time, over time, we've built out
- 20:02a ton of functionality of stuff that happens after capture.
- 20:05SO11, basic thing that you know, just happens kind of behind the
- 20:09scenes is like this storage and content management piece.
- 20:12You know, being able to manage 3D content, being able to export
- 20:15it into any format you want is still, you know, a big pain
- 20:17point in the industry. There's a lot of fragmentation
- 20:19with different file formats. So we invested in cloud
- 20:22infrastructure to be able to like back up your models to the
- 20:24cloud. You can log into our website and
- 20:26you could view them the same things that you view on your
- 20:28phone, you can view on the website, you can download them
- 20:31in, you know a dozen or so file formats.
- 20:34So we built a lot of content management features.
- 20:36We also started building in our own kind of first party editing
- 20:39tools like we have a crop tool, we have a rotate tool and
- 20:42there's a lot of other editing tools that we that we are
- 20:44planning to build. And so I think that that vision
- 20:48of like 3D capture itself is kind of too, too limiting
- 20:51because we are we're building a lot of non capture pieces.
- 20:54And now I would say you know our mission is really to democratize
- 20:573D creation as a whole. And I think we're starting where
- 21:02we have our beachhead as a company is on is is with capture
- 21:05or or creation creative work flows that start with capture.
- 21:08Often times they know they don't end with capture as you know you
- 21:11have to do a lot of post processing and and clean up to
- 21:13get them to something that's you know production ready and that's
- 21:16the sort of stuff that we want to build in into Polycam.
- 21:19But over time like going back to your early point about 3D
- 21:24creation being being hard to use, I think like most of the
- 21:27popular 3D creator tools today are you know they were built in
- 21:31in in the 90s, there were very advanced products with their
- 21:33times 90s or early 2000s. But there's been, there hasn't
- 21:37been that same modernization of those tools that we've seen with
- 21:40photo and video creation that are really you have easier to
- 21:44use user interfaces. There's usually powerful
- 21:46applications on web and mobile. There's a cloud component to it
- 21:50where it backs up and manages your content for you.
- 21:53There's like AI features that are coming out, you know, almost
- 21:55every, every week or month these days.
- 21:58And we're not seeing that innovation happen in these
- 22:00traditional 3D tools. So I think there's this big
- 22:03opportunity to kind of like build sort of the modern 3D
- 22:06creation stack from from the ground up and I think capture is
- 22:09going to be a really central part of that.
- 22:10But so we'll, you know, maybe maybe generative AI or so we'll
- 22:14work flows where you you start with it a model that you maybe
- 22:16you made in Maya, but then you want to apply some, you want to
- 22:19rig it and animate it and maybe you do that in polygamy.
- 22:24One more thing I want to ask you is about using 3D as a social
- 22:28medium. Now all the social medias are
- 22:30two D You send photos to your friend, you send videos.
- 22:33Do you see a future that 3D will become a primary element?
- 22:37I mean that was a topic that we were discussing with Paris.
- 22:39The other that slowly will be time that instead of sending
- 22:42photo to each other, we can send 3D to each other and we can view
- 22:45them quite. Spatial captures, I mean, that's
- 22:47what Apple was showing with the VR, right?
- 22:49Division Pro. So what what do you think about
- 22:52that? Actually I think if we're
- 22:55talking about like you know, long, long term, then yeah,
- 22:58absolutely. I think the like the thesis that
- 23:02basically you know the world is 3D and our brains perceive
- 23:06things in 3D and that's why like it's still so compelling to do
- 23:10somebody you know be out in the real world I think.
- 23:14And so therefore like eventually our a lot of our digital content
- 23:17that we consume will be 3. DI think that's as true of a
- 23:19thesis as it as it's ever been. And I think it's just you could
- 23:23have said that 30 years ago and sounded just as like a visionary
- 23:27or whatever. But I think the things that it
- 23:29turns out that it's just 3D is a lot more compute intensive.
- 23:32It's a lot harder like basically you know, getting a good like
- 23:35Gaussian splats I think are the closest that we have to
- 23:38something like a 3D photo. Like you look at it, you can
- 23:40turn it around and it and it looks, you know, very close to
- 23:43the real thing. But there's still still a ton of
- 23:46problems to solve. You can't capture people really
- 23:48well unless you have this like super expensive rig.
- 23:51Unless you capture every side of something, the up and the down,
- 23:54the back and the forth, like you'll end up with holes in your
- 23:56model. And so I think that like yes,
- 24:01that future is there, but we're still we're kind of early, early
- 24:04in that, early in that journey. We need better creation and we
- 24:07need better consumption. How promising is 3D Gaussian
- 24:12splats in your opinion as a technology?
- 24:17Yeah, I I mean, I think it's it's super promising.
- 24:20I mean, but there's the obvious reasons that it looks great,
- 24:23like we're seeing photorealism like we've never, never seen it
- 24:26before. I think even the quality is is
- 24:28better than some of the best nerves.
- 24:33But the thing that really kind of caught caught my eye again,
- 24:36kind of looking at it a little bit more critical, looking at
- 24:39nerves a little bit more critically and looking at its
- 24:40flats a little bit more critically was you know, one the
- 24:43performance is really good. You know, we can render these
- 24:46things in commodity hardware at A at a real time frame rate and
- 24:49that's something that you can't do, couldn't do with nerves and
- 24:52that's critical for the type of experiences like you're talking
- 24:55about sharing it with your friend on your iPhone.
- 24:59The the other thing was we started to see community
- 25:02integrations coming out so you could pull these things into
- 25:04other tools like Unity and Unreal and those things where
- 25:09people, because they're easier to render the nerfs, those
- 25:12things kind of just started popping up on their own.
- 25:14And that's a sign that like the ecosystem is kind of rallying
- 25:17around this, this format, this representation.
- 25:20Was that the primary reason that you guys delivered 3D Gaussian
- 25:24in such a speed so? 3D Gaussian paper was released
- 25:28back in August if I'm not mistaken during Seagraf and you
- 25:31guys delivered the product I think in the closest amount of
- 25:34time it was in two months. Yeah, We didn't expect it.
- 25:36We didn't. Surprise.
- 25:37Surprised when when you guys I think posted on Twitter, I
- 25:40messaged Barry's, I was like wow, what the hell?
- 25:42Like how did you guys, when did you guys started working on
- 25:45this? Was that the reason?
- 25:47Because what you mentioned initially about choosing the
- 25:49technology and for a company like Polycam, you know selecting
- 25:53a technology to be the part of your product on your app.
- 25:55It wouldn't be an easy decision. Was that the primary reason that
- 25:58you said, oh, I've seen 3D Gaussians, these are promising,
- 26:01let's go with them. And this is a long term future
- 26:03that we are betting on. Yeah, it was really like, I mean
- 26:08I saw the paper, I saw the things on Twitter.
- 26:10It was, it was impressive. But the thing that really pushed
- 26:14me to to want to adopt it was kind of seeing this community
- 26:16adoption because that's something we had never seen with
- 26:18nerfs, with nerfs. We had seen you know, really
- 26:21good results. We had seen a lot of excitement
- 26:22in the community, but it had it sort of hits this like the Super
- 26:25hardcore creators could use it because they can, you know, they
- 26:29have AGPU machine that can install the right version of
- 26:31CUDA, but beyond that like you can't just, you can't just pull
- 26:36them into Unity. So, so yeah, that was the thing
- 26:40that really swayed my mind. And it was actually from I've
- 26:43been watching it. But the point where I told our
- 26:45team that, hey, we're going to ship Gauss's last when we
- 26:47shipped, it was actually only a week.
- 26:49So we worked on it for a for a week.
- 26:51Oh my God. That was that was that was fast
- 26:54what you guys did there. Honestly we didn't expect it and
- 26:57what's really fascinating is that you guys are now upgrading
- 27:01your system. So I want to talk about the
- 27:03plans because was it today that you guys released the?
- 27:06Update yesterday, yesterday, yesterday.
- 27:08Yesterday. So can you tell us a little bit
- 27:10about the update because we tested it and we had less
- 27:13floaters, but I I wanted to go into more detail as to.
- 27:16Another important about the update beside the floater, what
- 27:18I love is the commercial use because from this, this, I think
- 27:22this update you can commercially use it.
- 27:24And for us because I mean we create content around it, a lot
- 27:26of people will reach out to us from entertainment industry,
- 27:29from real estate and they actually want to use this stuff.
- 27:31It's not only R&D. We do a lot of R&D every.
- 27:34Single video we create, people ask us.
- 27:36Metahumans, Google Maps in Unreal Engine 5 Gaussian split.
- 27:39They asked us can we use it commercially?
- 27:41Because most of our audience will want to use this for a
- 27:44project for a client and so they want to know and I think that's
- 27:47an important point. Thanks for bringing up SO
- 27:49Floaters commercial use. How did that come around?
- 27:52Because this was a pretty fast update.
- 27:53What? What else can we expect as well,
- 27:58right? Yeah.
- 27:58So when we launched, I mean another reason why we were able
- 28:00to do it so quickly is I mean shout out to the open source
- 28:03community, the original Gaussian splat paper which came out from
- 28:08from the research community. It was, it was a group primarily
- 28:10at INRIA, which is a computer science research group in in
- 28:13France. And they not only published this
- 28:16fantastic paper, they also released their released their
- 28:19code on GitHub to let other researchers use it and play
- 28:23around with it. And that was, you know, so we
- 28:26didn't have to recreate it from scratch.
- 28:27We were using you know their reference implementation which I
- 28:29think is probably the one that you used for your for your first
- 28:32experiment as well. It did come with this caveat
- 28:35though that it was for research purposes only and so it wasn't
- 28:38open for commercial use which is why we had this kind of non
- 28:40commercial disclaimer. And the one of the things things
- 28:44that changed is there's another open source project called Nerf
- 28:49Studio which has released a lot of released a lot of code for
- 28:54training nerfs and they came out with their own implementation of
- 28:57the paper that is available for commercial use.
- 29:00So we we started with with their version and then we started
- 29:03building on top of it and making these improvements to to remove
- 29:06floaters and speeding it up. So that was again you know
- 29:13working with the kind of the research and open source
- 29:15community is part of how you know companies like Polycam are
- 29:19able to to move fast because so many things are things are
- 29:22happening. But then I mean you don't want
- 29:25to just be using the open source version, you want to make
- 29:27something that's better you know that helps your your product
- 29:30stand out. And in our case, you know we
- 29:32have a really great in house computer vision and machine
- 29:35learning team and so we had been working like floaters were kind
- 29:39of the the most obvious. Effect of class.
- 29:42Most annoying, Most annoying, most annoying.
- 29:45Yeah, 'cause you might have the perfect model and there's just
- 29:47one floater standing in the middle of you and your model and
- 29:51it ruins the whole thing. Yeah.
- 29:54Or even worse. And you might have like a cloud
- 29:56of floaters and it's like this, like this this beautiful thing
- 29:58in there, but it just needs to like, you know, break it out
- 30:00from these floaters. So that was like when we were as
- 30:03as a computer vision team or like, looking like, hey, how
- 30:06could we what's how should we improve this?
- 30:09You know, there's a lot of things we could do.
- 30:10We could try to make it faster. We could try to make it more
- 30:12performance. But I think floaters were the
- 30:13first things that we had put in our crosshairs.
- 30:16And we we played out, played around with a number of ideas,
- 30:19and then we ended up finding one that that that worked really
- 30:22well. So I think you know from here
- 30:26like the thing that was impressive about Gaussian splats
- 30:29is that even the first version which still had these floaters,
- 30:32like it was still you know, production ready like it was, it
- 30:35was fast, it was good enough that people could start bringing
- 30:37it into these workflows. But it was the first version and
- 30:40you know Nurse Nurse, now we're on like version 100.
- 30:43You know, the first version came out in like 2018, gone through
- 30:46many iterations. And so I think the next couple
- 30:49years we'll see some big improvements for for Gaussian
- 30:52splat. Like we kind of the way that the
- 30:54AI research improvements happen is they're often happen around
- 30:59the major AI conferences. So there's like CDPR, there's
- 31:02SIGGRAPH and I think we're at the end of the year.
- 31:04So there's no more conferences for the rest of this year.
- 31:06But like I don't know what the first conference of next year,
- 31:09but we'll probably that's when we'll start to see lots of waves
- 31:11like the research community, everyone took note of Gaussian
- 31:14splat during SIGGRAPH. Tons of research labs started
- 31:17working on how do I make splats better and then we'll start to
- 31:19see those first like in from the research community.
- 31:22The other improvements that other research labs figure out
- 31:24come out, you know, probably probably early next year, so so.
- 31:28I I I want to talk about one thing.
- 31:30So you guys updated the Gaussian splats on the website.
- 31:33So now for anybody that tries, they're going to end up with
- 31:35less floaters. We have the commercial license.
- 31:38For people who have watched our videos or they've trained their
- 31:41own Gaussian splats, we know that we get different iterations
- 31:44and of course the higher iteration I believe it's 30,000
- 31:47produces and yields a better looking result.
- 31:49I want to know for people who are making a decision.
- 31:51Of course when you go to Polycam, the advantage you have,
- 31:54not only is it fast, you can upload multiple and you you know
- 31:57you can basically just use this if you've never had any
- 32:00experience training your own or don't have the graphic card
- 32:03capability to train your own. I want to know if there's a
- 32:05difference in the way you guys created your algorithm because
- 32:09the fact that it has less floaters.
- 32:11Is it related to higher iterations or no?
- 32:13This is the Polycam specific algorithm.
- 32:16And so if people are testing it with Polycam, they might
- 32:18actually end up getting a different result, a cleaner
- 32:20result eventually, then maybe training it themselves down the
- 32:24line, right? Yeah, no, they will.
- 32:26Absolutely. It's not just more iterations.
- 32:28We made some some basic changes to the algorithm that's not
- 32:31available in any of the other implementations out there right
- 32:34now and so implement. I mean iterations.
- 32:36I wouldn't get overly fixated on iterations because if you
- 32:40improve your algorithm then you can get a better result with
- 32:42fewer iterations like that. That is that is the ultimate
- 32:44goal. So I think right now we use
- 32:4615,000 iterations, but I think probably for most, the majority
- 32:51of data data sets that polycams pipeline will produce something
- 32:54that's better than the default pipeline for for 30,000
- 32:58iterations and so, so yeah, it's, it's, it's, it's both.
- 33:03I think it's OK. That's one that's all around
- 33:07now. And what?
- 33:08About the future of Gaussians, because this is something that
- 33:10neon farrows are really interested in.
- 33:12I saw that you tweeted about in the future being able to animate
- 33:17this Gaussians the same way we are.
- 33:18You know we are connecting a skeleton to any of these blobs.
- 33:21What do you think about having mesh and polygons with 3D
- 33:25Gaussians? Because that's the most
- 33:27commented. Post where there are videos that
- 33:31is there a future for. I mean, at the moment you always
- 33:33say no, there is no mesh, there are point clouds and if you
- 33:36bring it to a software like Unreal, it's.
- 33:38Niagara particles, Niagara particles, There's a new plug
- 33:40in. We have to test it out.
- 33:41Apparently doesn't use Niagara particles, but that has yet to
- 33:44be tested. But up until then I think for
- 33:47people and for us yeah the question that fired asked I
- 33:49think is is is very popular yeah so that's something what we're
- 33:56working on internally as well there isn't I wouldn't say
- 33:59there's there are definitely ways to get a mesh out but you
- 34:03can use you can use an algorithm called marching cubes which is
- 34:07an algorithm from like the 80s to try to get a mesh from point
- 34:11clouds something well from point clouds or for as a sine distance
- 34:15field but or maybe you've heard of TSDF those are used those are
- 34:19used but it's something you have if you have something that's
- 34:22sort of like you can if you can turn it into a voxel
- 34:24representation which you can then you can extract a mesh but
- 34:28you you won't get a mesh the mesh won't be super smooth and
- 34:31it won't have all these like it won't look nearly as good as
- 34:35that's the thing when it's in Gaussian.
- 34:36Yes, that's the thing. That's what we were thinking.
- 34:38Because I remember I I searched on YouTube and there is already
- 34:41this application that turns point clouds, PLY files into
- 34:45polygons. Well then the quality, all of
- 34:47the benefits of Gaussian splats are not going to be there at the
- 34:50moment to do it, which is why we asked the questions.
- 34:53People just want to know, can I, for example, scan a person
- 34:55eventually down the line and it's going to look as realistic
- 34:59and then be able to attach an armature to it and start moving
- 35:02it around. And that's why I think a lot of
- 35:03people are. Having these questions as to
- 35:05when it's happening. So it's it's a possibility.
- 35:08You would say potentially it it is I, you know and honestly I've
- 35:13kind of shifted back and forth on this I think for a long time,
- 35:17part of and again part of my reluctance of not adopting
- 35:19nerves was because you need to get a nice mesh out of it.
- 35:22And I would say that like oh, the whole computer graphics
- 35:25industry runs on triangles. Like if you can't get it into
- 35:27triangles, it's not going to get adopted.
- 35:29And I I mean one thing that what caused me to walk that back
- 35:34again was just seeing people build these plug insurance for
- 35:40for Gaussian splats. And I think there's a question
- 35:42of like is there room for another another representation
- 35:45like meshes are the most common one, but we also have point
- 35:48clouds and those have been around for a long time, they're
- 35:50still used. We also have things like you
- 35:52know, volumetric things that are used for for fog and for
- 35:55liquids, which is not exactly a mesh either.
- 35:58And so I honestly, I think getting a mesh from, like I say,
- 36:04a scan of a human that has the strands of hair, that that's a
- 36:08very, very hard thing. Like, there's a lot of computer
- 36:10graphics students that do their PhD on hair.
- 36:14You know how do you how do you how do you how do you how do you
- 36:17make how do you make hair hair look good and it's really hard
- 36:20like a hair and a lot of you know characters in video games
- 36:24it looks crappy and. To this day, 2023, to this day,
- 36:27we're not getting it right. Sometimes I understand, yeah.
- 36:31And so the splats, I mean it's granted it's not animated hair,
- 36:33it's not going to like bounce if the character moves but it for
- 36:36the for some of the ones if you go to our like Polycam slash
- 36:40splat page, there's some cases where the hair just looks looks
- 36:44excellent and so is. That the one with the lady, I
- 36:48think. Did you capture that with the
- 36:49lady and the basically I think there's a bunch of cameras
- 36:53around her. I think that's one of the top
- 36:54ones. Her hair looks fabulous in that.
- 36:57Yeah, no, I didn't. That was that was a community
- 37:00upload. Honestly, there were, I was
- 37:01surprised how many uploads there were of characters in these kind
- 37:05of photogrammetry rigs. I think it speaks to there being
- 37:07this really need of like really big need of like how to capture
- 37:10people even. Because because The thing is,
- 37:13even if you had one of these multi $1,000,000 rigs using
- 37:17photogrammetry, you would still end up with you know, the hair
- 37:19looking weird and you know stuff like you know the little the
- 37:23little lights on the eyes that kind of move around, the
- 37:26reflections on the eyes that change as you as you move around
- 37:30a person. But they make that person look
- 37:31alive rather than like a statue. So I think that there will
- 37:36continue to be tension between like I think things will
- 37:43continue to look better in the native Gaussians flat
- 37:45representation and we'll have to see what the research community
- 37:50does with it. I think it's, I think it's
- 37:52totally possible to animate them I think but doing things like
- 37:55relighting them I think will fully relighting them, I think
- 38:00that that will be that will be a challenge.
- 38:03But yeah that you know it's a it's an amazing time to be in
- 38:08the computer graphics industry right, because we're surprised
- 38:10every, every, like every month by new stuff that comes out.
- 38:12So I'm I'm reluctant to make strong bets, but I would say
- 38:16it's there's not there's not an easy answer, an easy solution.
- 38:19But totally understandable, and I I kind of know what question
- 38:23you want to get into and I'm going to lay the grounds for it.
- 38:25But I just want to say I also appreciate the fact that you
- 38:28said this might just be another 3D representation and might not
- 38:31have to turn into polygons. That's something for people to
- 38:34consider and ourselves. We want to talk a little bit
- 38:38about AI. And I'm just going.
- 38:39To lay the grounds by saying you mentioned within the 360
- 38:44application of Polycam, you guys take photos right of what the
- 38:48basically the wide angle, you guys use the wide angle lens and
- 38:50then you fill the top and bottom with stable diffusion.
- 38:54So we had one question related to that.
- 38:56So for 3D Gaussians, do you think it's possible that we can
- 38:59start incorporating AI in a way that we can fill?
- 39:03The missing frames and angles that vary now raw input.
- 39:07So now for example if I take a photo from three quarter of a
- 39:11building, I will have it perfect, but I can't go around
- 39:14it fully because that's that's frames were missing in my
- 39:17capture. Do you think with AI we can
- 39:19reveal all of those in the future?
- 39:22Yeah, I think that's one of the, like the most most important
- 39:25unsolved problems in this field of 3D reconstruction.
- 39:28You can totally see it. How like an AI could fill that
- 39:30in because they can fill in these images super well and
- 39:34there's impressive stuff with video with with 3D.
- 39:36You know one of the bottlenecks is you need a lot of all of
- 39:39these in in the 3D case these the stable diffusion that can
- 39:43fill in big holes in the in the image it's it can do that
- 39:46because it's seen every image on the Internet, right.
- 39:48Almost. There's billions of billions of
- 39:50images and there's not like billions of 3D models certainly
- 39:54not billions of 3D models that are just out there on the
- 39:56Internet. But yeah, I something I think
- 40:00it's. Like the most, maybe the most
- 40:02important problem. It's one that we're that we're
- 40:04thinking about you want to work on internally.
- 40:06It's one that I think a lot of, you know, research labs are are
- 40:09working on too. Do you think we have hardware
- 40:12challenge for that as well? Because I saw that you guys
- 40:14posted on Twitter, when you guys released the update, there were
- 40:18heavy usage of, you know all these programs on your
- 40:21application. I can imagine that if you add AI
- 40:23to this, it would just go haywire.
- 40:26Is that is that true? Like the hardware would be also
- 40:28a challenge if you start incorporating AI.
- 40:33I I don't think it. I don't think it's a blocker no.
- 40:35I mean we're already using you know GP US for compute the
- 40:38Gaussian splats. It's not strictly speaking AI
- 40:40but it it it uses a lot of similar sub like sub algorithms
- 40:45that are used to train neural networks.
- 40:48I think this problem will be solved.
- 40:49I'm very lucky. I think it needs to be solved.
- 40:51It's the key to making 3D capture easier, right?
- 40:55I think the ideal 3D capture experience is where it's almost
- 40:58as easy as taking a photo. You know, you don't have to do
- 41:0033 circles around something. You know, the the ultimate thing
- 41:05is you could take take a photo of me.
- 41:07And then like if you took a photo, if you took this photo of
- 41:10me and you handed it to a really good 3D artist and you said,
- 41:13hey, turn this into a 3D scene, they could, they could do it
- 41:16because they understand that like the stripes on my shirt
- 41:18will continue to the back of my shirt.
- 41:21They can they can guess what the rest of my the room looks like
- 41:24and it might not be, it's not going to they don't know exactly
- 41:27what you know are the books on the bookshelf behind my head,
- 41:30but they could put in some reasonable books, right.
- 41:32And I think that we will have AI that that can that can do that
- 41:36sort of interpolation and translation and that'll be huge
- 41:39for that is one of the things that's needed to like your
- 41:42question of like how do we make 3D mass market.
- 41:45I think that's one of the missing pieces and it will yeah.
- 41:50I mean we'll, I think we'll be certainly over the next 5 to 10
- 41:53years we'll be able to kind of see that, see that problem get
- 41:56solved and then enjoy the enjoy the fruits of it.
- 41:59I think I think this also if if the AI is coming in at the input
- 42:03layer where it's fixing the images, I think it also has to
- 42:06be consistent right? So if it's filling in let's say
- 42:08for 50 images that is going around the house, it has to have
- 42:13the same consistent texture on the same part of the house so it
- 42:16looks. You know, it looks right when it
- 42:19comes out as 3D unless it's not coming in that layer like you
- 42:22mentioned. If it's in the 3D layer, I'm I'm
- 42:24going to see what happens. But I'm hoping this happens
- 42:26because a lot of our captures we have the, we have everything
- 42:29perfectly there and we're taking multiple angles in Unreal Engine
- 42:33and then we're like, I just wish I had this extra angle that I
- 42:35couldn't get when I was there. So that's why we had to ask this
- 42:38question. I want to just take it back to
- 42:41the very beginning of our conversation.
- 42:43You were talking about AR cloud and I didn't even know the the
- 42:46term AR cloud. Do you think that's still
- 42:48happening? Do you think that's going to
- 42:50happen now that Apple came into the whole game of, you know,
- 42:52having Vision Pro on your head? Of course the form factor needs
- 42:55to get smaller, but where do you see that going and are you going
- 42:58to be a part of that when it happens?
- 43:04Yeah, I think that I think that we'll still be surprised by the
- 43:10killer. Killer apps.
- 43:11I think the killer applications of AR and AR glasses, I think
- 43:15that there was like people saw Pokémon Go take off and then
- 43:18they were like OK, this is the killer use case.
- 43:19We're going to have like dinosaurs, you know running
- 43:22around our running around the streets and stuff like that.
- 43:25And I I think that's I think that's like not going to be that
- 43:30that could be like a fad that people are excited about and
- 43:33like you know has a has a peak and and a bust.
- 43:35But I think enduring use cases will be ones that deliver kind
- 43:39of more value. So I think in the AR cloud
- 43:41situation, you know, there could be things that actually layer
- 43:44meaningful content on top of the real world.
- 43:46Like if you're in a museum and there's like a, you know, a
- 43:48plaque to the side that explains what you're looking at.
- 43:51Or things that give you directions like things that give
- 43:53you valuable information at the right time in a sort of
- 43:56contextual way. Well I mean in that sense like
- 43:58using using Google Maps is already kind of like an AR
- 44:02feature if you will because you're navigating it on the real
- 44:04world and it's telling telling you giving you digital
- 44:06information that's contextual and related using GPS contextual
- 44:10information your your immediate surroundings and so but I think
- 44:16there's yeah there's a there's a lot to figure out.
- 44:18The main way I think that we will be involved is that like a
- 44:20lot of those AR experiences will need to have 3D content to to
- 44:26realize them. And I think our our mission is
- 44:28to write really build tools that empower the 3D creators that
- 44:31need 3D models or they need back 3D environments or 360 photos to
- 44:37make their to make their AR and VR experiences.
- 44:40And we'd love to be, you know, the tool that's powering a lot
- 44:44of that creation. Speaking of empowering the
- 44:51creators, are we going to see editing tools for Gaussians on
- 44:54Polycam in the future? Is it a part of the road map or
- 44:57no, I just. Want to drag on top of the
- 45:00Gaussians? Just delete the ones that I
- 45:02don't want. That's that's my dream come
- 45:04true. We don't have anything to
- 45:06announce yet but yeah I think it's I'm in the same boat you
- 45:11know I want to I want to edit the Gaussians.
- 45:13There's not there's not great tools out there unlike you know
- 45:16we we we launched you know photogrammetry but there are
- 45:19already great tools for cleaning up a a mesh like we can bring
- 45:22some of those into Polycam to make it a little bit more
- 45:24streamlined for people but it doesn't change what people can
- 45:27do. But right now yeah there's
- 45:28there's just like there's this clear hole in the market where
- 45:32you have your splat you want to clean it up.
- 45:34There's not not great tools for doing that.
- 45:36Yeah. And I just have one final
- 45:39question that I want to ask. I think Chris has been into
- 45:42multiple market cycles and he has seen, I mean AR Cloud, I
- 45:46think NFDS and Web Tree. And now with AII want to know as
- 45:50a CEO of a startup of a company, how would you advise other
- 45:54entrepreneurs to detect something that is an actually a
- 45:58trend like what or something that is added with a critical
- 46:01eye on you? Yes, exactly what.
- 46:03You said yeah. How would you?
- 46:06If someone asks you that, oh Oh my God.
- 46:07You know, nowadays it's very hard to be an entrepreneur,
- 46:10because today I wake up, there's a new tool that might kill my
- 46:13app. Tomorrow I can wake up in
- 46:14another. One, and I'm also competing with
- 46:15the big guys, right? They always have more money
- 46:17because I I know people have this mentality, especially when
- 46:20they're small. But you already surpassed all of
- 46:23that, right? What would you give as an
- 46:25advice? I I love that question.
- 46:27You know, the first time I went into AR with this AR shopping, I
- 46:30knew nothing about shopping and nothing about AR.
- 46:32The second time you know I've been working at an AR company
- 46:35that was building this 3D capture tool display land and
- 46:38they were that tool was serving 3D creators.
- 46:40So I already had you know a a good understanding of that
- 46:44market and the the applications for for 3D capture.
- 46:46And so when I started Polycam, I came into it with that good
- 46:50understanding of of the market. And I think so I think working
- 46:54on something that where you where you know an industry, I
- 46:56mean you can always expand out outside of that if you're if
- 46:59you're growing a successful company it will expand.
- 47:01But starting with something where you understand the pain
- 47:04points that's still something in AI can't automate for you.
- 47:06You know I can't tell you that like oh is there there's this
- 47:09specific need in the market. So for your audience, you know
- 47:11it might be people knowing certain certain problems in the
- 47:15computer graphics industry or certain pain points like when
- 47:18they're they're trying to do something in unreal and it's
- 47:19just like oh this spam hasn't been solved.
- 47:22But if they're if they're a coder or they work with a coder,
- 47:24then maybe they could build, you know, a small tool for for
- 47:27solving that problem. I mean, I think one people
- 47:29should always be a little bit suspicious of hype when
- 47:32something is. If something is blowing up on on
- 47:34Twitter, it it either means that it's like it's a fad or it means
- 47:38that like that cat's already out of the bag and and that you'd be
- 47:42too late if you were. If you wanted to jump on that
- 47:44bandwagon, you need to you need to trust yourself, your your
- 47:47instincts more than trying to copy what other people are
- 47:50doing, 'cause like I said, I got interested in 3D capture in
- 47:5220/16/2017. Nobody interested in 3D capture
- 47:56at that point so so I would say like you know if you if you have
- 48:00an an interest and you think something that you're kind of
- 48:04drawn to. I wouldn't.
- 48:05I wouldn't be discouraged if it if it's not like the the thing
- 48:11on on Twitter now you know if it if it's something if there's
- 48:13some real value there then it probably will be at some point.
- 48:17But, but most of the people that are building successful
- 48:20companies, they were they were working on that thing before it
- 48:22was cool. And if they were successful then
- 48:24maybe they helped to make it make it cool.
- 48:26And there's I think it I also think it does take practice and
- 48:30you you can't be afraid of of failure.
- 48:32Like this is my kind of polycam's kind of my third time,
- 48:36third time's the charm. Third, third time around.
- 48:39And so I think people can also get into this like analysis
- 48:43paralysis where they have some idea but they could see all the
- 48:46ways that would fail. And so they never kind of go go
- 48:48in and I think you have to go all in to to to to learn the
- 48:51things that you don't know. You don't know what you don't
- 48:53know. And the best way to get that is
- 48:55to kind of is, is to get to get experience and and get your
- 49:00hands dirty and feel the pain of failure and then like we were
- 49:04talking at the at the very beginning, you know that in turn
- 49:07can lead to the, the, the next thing.
- 49:12There you have it guys. That is the best way to conclude
- 49:16all of this. Motivational, but more so than
- 49:19motivational inspirational message from Chris.
- 49:21Thank you so much. For joining us today, man, we
- 49:24had just the best time talking to you and we're even more
- 49:28excited to see what you guys work on, especially now that we
- 49:30know some of the stuff you guys are working on are going.
- 49:32To be holding you accountable for all this.
- 49:36Keeping it. So we're looking forward to all
- 49:39of those updates, Chris. No, it's my pleasure.
- 49:41Thanks for having me on and talk to you again.
- 49:43No. Worries, man.
- 49:44Thank you so much. Will do, man.
- 49:46And thank you to everybody listening and watching.
- 49:49If you guys haven't tried out Polycam to this day, come on,
- 49:52we've already made so many tutorials going give it a try.
- 49:54Especially 3D Gaussian splats. That's the one.
- 49:56You should be creating now you can use it with a video.
- 49:59You don't even. You don't even need to take
- 50:00photos. Just record a video.
- 50:02See the result. You'll be amazed.
- 50:04Easy, man. All right, we'll see you guys on
- 50:05the next episode. Until then, Ciao.