Nicolas Burrus

3D computer vision at . Also created RGBDemo, Skanect, DaltonLens, nbplot.

[Nicolas Burrus @nburrus]

Can be bought DRM-free from the author's webpage https://t.co/tYUJYU0VK8

[Nicolas Burrus @nburrus]

This book "Enshittification" is a must-read to understand what we’re living through. https://t.co/SE4GjUFqpo

[Nicolas Burrus @nburrus]

@metro_madrid un metro cada 9 min el finde en la línea 5, porque tan pocos? Apretados como sardinas todo el trayecto con dos niñas, intentamos usar transporte público al máximo pero la próxima será con coche :-(

[Nicolas Burrus @nburrus]

@JeromeRevaud @luxonis very impressive, great work! Definitely need to integrate this in stereodemo when the code will be out :)

[Nicolas Burrus @nburrus]

I wanted to see how the latest stereo approaches perform in the wild with my @luxonis OAK-D Lite camera, so I made stereodemo. It compares 4 deep learning methods + OpenCV baselines on files or OAK-D streams. Just one "pip install" away: https://t.co/4cO4ONoj0Q

[Arcturus Vision @arcturusvision]

We've been busy! Introducing a new Gaussian Splatting engine that's lean and (well above the) mean.

🥇 Truck scene: 68 sec / PSNR 24.19
💼 Commercial-friendly, pure C++/CUDA

Interesting? Let’s talk!

#gaussiansplatting #ai

[Nicolas Burrus @nburrus]

@LaFranceOSi Bonjour, y'aura-t-il bientôt un nouveau formulaire sur votre site ? Le formulaire sur https://t.co/bkB4d7OJWo est toujours pour le troisième trimestre 2022/23.

[LAION @laion_ai]

We launche a petition to democratize AI research by establishing an international, publicly funded supercomputing facility equipped with 100,000 state-of-the-art AI accelerators to train open source foundation models.
https://t.co/jbkQnlEHYH
https://t.co/agmh1Ptjky

[Jeff Powers @jrpowers]

Teleporting into a NeRF in VR. Built on top of #instantnerf. Running on an RTX3090.

Anyone interested in this (for practical or fun reasons)?

#nerf #vr

[Carlos Campos @carlos_cam_mar]

Thrilled to shared that I was awarded with the #GTRob award for the best thesis in robotics in Spain! Thanks to #UniZar for having the opportunity to work and learn from some of the best SLAM researchers. Now let's keep working on real SLAM, new exciting times in @arcturusvision

[Angjoo Kanazawa @akanazawa]

Releasing 🌟nerfstudio🌟, a plug-and-play python library to easily create your own NeRFs!
@nerfstudioteam is a contributor friendly open-source repo with a realtime web viewer that makes it easy to make cool videos 📽️
https://t.co/zw0ZPqIZ4c
https://t.co/NZM8GPjNrF #nerfstudio 1/

[Nicolas Burrus @nburrus]

Having great fun with @ElevenVR , feels very realistic! It's also interesting how you can get hints about the personality of other players even without video nor audio, just from corporal language playing table tennis.

[Nicolas Burrus @nburrus]

@grbradsk @luxonis Yes indeed I did some speed/memory comparison, and while I didn't try to run any of them on device, I'm quite hopeful for several of them once optimized and configured with a speed-oriented trade-off. Would be very cool to see!

[Nicolas Burrus @nburrus]

@luxonis Just added DistDepth (CVPR 2022) https://t.co/UwLbkx0SAs . It is a monocular method and thus only uses the left image, but still interesting to see how it compares to the stereo methods. `pip install --upgrade stereodemo` to try it.

[Nicolas Burrus @nburrus]

@luxonis Thanks, glad you found it useful. Being able to run this kind of approaches on device would be awesome, good luck optimizing the models!

[Nicolas Burrus @nburrus]

@luxonis Wrote a small blog post comparing these algorithms more in depth, with speed/memory measurements and longer comparison videos: https://t.co/tWnyTQHUBJ

[Nicolas Burrus @nburrus]

@luxonis Just add STereo TRansformers too (ICCV 2021). pip install --upgrade stereodemo && stereodemo --oak to compare them on live images

[Nicolas Burrus @nburrus]

Thanks to @PINTO03091 and @ibai_gorordo for collecting/converting a bunch of pretrained models and making them easy to use!

[Nicolas Burrus @nburrus]

Overall the latest deep learning results are pretty impressive on things like blank walls, even if they were not trained on that. In particular I found RAFT-Stereo (3DV 2021) to have a nice perf/quality tradeoff with its real-time settings.