Papers of the day   All papers

NASA Neural Articulated Shape Approximation

Comments

Geoffrey Hinton: This is a really neat way of using neural networks to get rid of polygonal meshes. The interaction of computer graphics with neural networks is really exciting.

10 replies, 1028 likes


Andrea Tagliasacchi: NASA: Neural Articulated Shape Approximation Representing characters and deformable shapes without polygonal meshes for 3D perception, games, ... https://arxiv.org/pdf/1912.03207.pdf https://t.co/nlkyIRJ7NU

3 replies, 350 likes


Andrea Tagliasacchi: Do you like implicit functions to model 3D geometry? We show that "Neural Articulated Shape Approximation" improves the approximation power by over 50% ABSOLUTE!! Fast-forward of our #ECCV paper, full talk, additional material, and code comings soon. https://arxiv.org/abs/1912.03207 https://t.co/7tnJ8Jw25C

2 replies, 235 likes


Andrea Tagliasacchi: Double-blind is great, so why cannot #arXiv add an "anonymous till published" mode? The authors could then be revealed only once the paper is full peer-reviewed (or after a cutoff date), and you could still cite them, thanks via https://arxiv.org/abs/1912.03207

1 replies, 59 likes


The TL;DR of 3D Deep Learning.: Deng et al. "NASA: Neural Articulated Shape Approximation" #ECCV2020 A representation of articulated deformable objects using neural indicator functions conditioned on pose useful for 3D tracking, reconstruction, animation... https://arxiv.org/abs/1912.03207 https://t.co/mwtl0TH07i

0 replies, 26 likes


Gerard Pons-Moll: Check out NASA (ECCV) -- an implicit function controllable with articulated pose.

0 replies, 25 likes


Shivon Zilis: Fitting that NASA outputs a spacesuit. Fascinating work!

0 replies, 16 likes


Nima Ghorbani: It is a wonderful, and humbling feeling to walk back home waiting to read two papers in your backpack during the holidays, which build on your previous work and go beyound what you could have achieved before! VIBE: https://arxiv.org/abs/1912.05656 and NASA: https://arxiv.org/abs/1912.03207

1 replies, 3 likes


Ryan Schmidt: Seems cool even though it doesn’t have any meshes.

0 replies, 3 likes


Andrea Tagliasacchi: @gabrielpeyre @GerardPonsMoll1 And this has applications in vision/geometry (registration of deformable bodies) where this convolution can be replaced by an expectation → https://arxiv.org/pdf/1912.03207.pdf https://t.co/CcUZFmweFK

0 replies, 3 likes


Daisuke Okanohara: NASA is the first neural implicit rig and can represent the shape of articulated objects. Each part is represented by an implicit function with a position and parameters being transformed by the pose. The whole is represented by their union (max op.). https://arxiv.org/abs/1912.03207

0 replies, 2 likes


Content

Found on Apr 24 2020 at https://arxiv.org/pdf/1912.03207.pdf

PDF content of a computer science paper: NASA Neural Articulated Shape Approximation