Hacker News new | past | comments | ask | show | jobs | submit | dulyanov's comments login

Hey HN,

I'm Dmitry, one of the founders of in3D.

With in3D app [0], anyone can create their realistic 3D avatar with just a phone camera within 30 seconds. Once scanned, users can animate their avatars, make funny videos in AR, try-on digital looks, and play as themselves in GTA V, Second Life, and VRChat. Or order custom jeans at [1].

We believe that sooner or later digital copies of humans will become useful for many people. So we are building tech for digitization. There are two goals we try to achieve: 1) quality 2) accessibility. For the latter we try to design the scanning process to be as convenient as possible, i.e. we try not to require too much from the user, help users during the scanning process and fix any issues such as lighting conditions transparently to the user while building the avatar. For the quality -- MetaHuman quality would be great :)

One of the major challenges here is the absence of the datasets with 3D humans scanned in real environments. So it's not possible to create an end-to-end DL-based solution that works well enough. We managed to use FaceID sensor of the iPhone to build avatars without using much DL. And now we have a dataset of ~70k scans that we use internally to improve the algorithms.

We are also building some applications for the 3D avatars: our app [0] that is now more like a demo of the technology rather than a product, the Virtual Fitting Room for desktop [3] that we launched with digital fashion house Replicant [4]. We have helped Softbank scan people for World Gymnastics Championships [5] and celebrities for their game [6]. Soon, the game will have scanning embedded inside of the game app.

Would love to hear your thoughts and feedback! You can try the app here https://apple.co/3vnUf3H (3 day free trial), Android version is coming in January.

[0] https://apple.co/3vnUf3H

[1] https://unspun.io/

[2] https://trai.in3d.io/

[3] https://www.youtube.com/watch?v=Xl0ffFJW8cU

[4] https://www.instagram.com/replicant.fashion/

[5] https://in3d.io/blog_page/in3d-partners-with-softbank-to-cre...

[6] https://www.youtube.com/watch?v=E3QZHi_rjy4


Hey HN,

I'm Dmitry, one of the founders of in3D.

With in3D app [0], anyone can create their realistic 3D avatar with just a phone camera within 30 seconds. Once scanned, users can animate their avatars, make funny videos in AR, try-on digital looks and play as themselves in GTA V, Second Life, and VRChat. Or order custom jeans at [1].

From the beginning, we wanted to get scanning UX as easy as possible so anyone can do this. We started off building a Virtual Fitting Room -- try-before-you-buy experience powered by a 3D scanning tech that we believed we could develop for smartphones. When we got the scanning working we realized that once a person is digitized the possibilities are much wider than just visualizing clothing. That's what we showcasing in the app now. Turning this in a product is still yet work to be done.

in3D exists for about 2 years now and it's been a really tough journey so far. My background is 100% technical -- I love coding, breaking/fixing stuff, did PhD about neural nets for image synthesis. Yet, by the end of PhD program, I knew I don't want to pursue an academic path and want to make something people would use. The problem was -- I had no idea how to start. Moscow is not Silicon Valley, people don't talk about startups on the streets and it was really hard to connect to people who built successful startups. I was lucky that in 2019 CVPR conference was in LA, so I booked a flight to SF and spent both CVPR and the week in SF asking how to build a company. I remember I asked a guy "Can you explain me how to do business development?" and when he responded "You just go and talk to people and businesses" and I thought well, it doesn't help much, I thought he would give me a to-do list or an algorithm. But we got lucky, we were accepted by Alchemist Accelerator, so we had a fantastic reason to move to SF. We learned a lot. The demo day though was in the beginning of February 2020 and we ended up raising only $50k before Covid. The tech did not work well at that time and we had no product. Luckily, we found some consulting work, got some profit from currency exchange, and managed to survive the hard times and raise more 8 months after. It became a bit less stressful since. Yet I understand better and better why investors are not excited about first-time founders -- we make mistakes every freaking day no matter how much we talk to experienced mentors.

Another interesting bit is how I met business co-founder. We connected during Covid, I was in US, he was in Israel. We've met in person only in a year from the moment he had shares assigned. Might sound blunt, but it's just been very natural -- he started contributing to in3D without asking permission or waiting on formal agreements. This might be the only way the co-founding relationships can start, some magic should happen.

We have also launched on ProductHunt today [2], would appreciate your support!

[0] https://apple.co/3vnUf3H

[1] https://unspun.io/

[2] https://www.producthunt.com/posts/in3d

[3] https://www.youtube.com/watch?v=Xl0ffFJW8cU


We have recently built https://www.co-train.space/ to let people exercise online with Youtube video. See discussion here: https://news.ycombinator.com/item?id=22890595


Youtube sync should work for lots of people.

We tested video calls in a group of 8, it was working well. I heard that having around 35 people in one room can make it unstable.


No, there is only one Host who governs the video, everyone else has no ability to change the timestamp or stop the video.


Oh, I think I did not test this. Completely forgot I have YouTube subscription. I guess it should be possible to detect when and starts and handle it. But sudden pauses due to ads can ruin UX especially for fitness...


This was posted by my wife :)


That could be fun! I saw Sway app doing that for dance movements, but in fitness such tech for sure could be more useful.


Glad you liked it!


Technically syncing is easy, websocket + Youtube API. The biggest issue for UX though is that a user should click on the video at least once. Otherwise, Firefox will block sync commands.

The latency can be 3-5 seconds for the first sync, when the video is initialized and unbuffered. For the next syncs its < 1 second.


Interesting as I would of thought fitness video's would need tighter latency due to the beat, but then, that can actually work in your favour as the beat will be way less than a second, so could always sync upon that for what would appear to be perfect sync, yet 1-3 beats offset beat out of sync and yet the illusion of perfect sync.

With workout videos usually 120-140 BPM, sub second, let's say 1 second for ease on 120 BPM would be 2 beats per second and so a sync of 2 beat offset would be spot on. Now a 140 BPM would be 3 beats, but you get the idea. By syncing the beats with an offset to cater for the latency and the BPM, you can give the illusion of perfect video sync. More so when doing repetitive motion to that beat, just seems like it would work really well. Certainly makes management of latency a much more easier task and allow for buffering as well depending upon how many beats you want to offset, so could handle latency spikes and brief drops. Which again, makes the things easier when you can control and fix latency , though as you can't do that, planning it out of the equation is always a treat. Food for thought perhaps.

But it is your baby, you pick the formula you want to feed it. But hope you do well whatever direction as certainly has potential in many avenues. Remote rave anybody :).


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: