Chuang Gan: #ICLR22 Is it possible to create dig...| Robocrunch AI

#ICLR22 Is it possible to create digital twins by just watching videos alone? Please check out our RISP framework that allows solving this challenging inverse problem by integrating deep representation learning with differentiable simulation and rendering! pic.twitter.com/KpSMVa2JvI
— Chuang Gan (@gan_chuang) May 6, 2022

Yes, we use RGBD and multi-view 3D here, but we have similar experiments with single image 2D networks @anagh_malik https://t.co/d9zlPOxtdU. I believe unsupervised self-similarity gets you nearly all the way; just the icing/cherry from supervision or reinforcement needed on top.

Shared by Andrew Davison at 3/28/2022

Video Frame Interpolation Transformer abs: https://t.co/JtQik8ahVt

Shared by AK at 11/30/2021

Adding depth warping to my neural style transfer pipeline, like all the cool kids do with diffusion/vqgan videos. Part of an upcoming collaboration with @va2rosa!

Shared by Gene Kogan at 3/6/2022

Dear lazyweb, Is there something analogous to BERT/wav2vec2 but for video rather than text/audio data? So a model trained on video clips in a self-supervised way without any labels?

Shared by Grzegorz Chrupała 🇪🇺🇺🇦 at 2/23/2022

Multi-View Consistent Generative Adversarial Networks for 3D-aware Image Synthesis abs: https://t.co/ja83CLkool

Shared by AK at 4/13/2022

Excited to share new work on pre-training a single reusable representation for many robot manipulation domains & tasks R3M pre-trains on human videos & then learns w/ only ~10 min of demos of downstream task https://t.co/aXCYePVH0H w @SurajNair_1 @aravindr93 @Vikashplus A Gupta

Shared by Chelsea Finn at 3/24/2022

768