Gerard Pons-Moll's banner

Gerard Pons-Moll

@GerardPonsMoll1 • 6,359 subscribers

Professor of Computer Science at the University of Tübingen. Machine Learning, Graphics and 3D Computer Vision enthusiast.

Shorts

Can you generate full body motion from **just** a head mounted camera? No, of course that's impossible! ... or is it? Check out HDM^2: Work by Vladimir Guzov and colleagues while interning at Meta:

Can you generate full body motion from just a head mounted camera? No, of course that's impossible! ... or is it? Check out HDM^2: Work by Vladimir Guzov and colleagues while interning at Meta:

11,591 views

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

Capturing every combination of human-object interaction is not feasible. We need compositionally. Most interactions are local. A hand holds a cup, a chair supports the pelvis, while much of the body remains free. COSMI, by Daniel Escandar Ilya Petrov builds on this observation to compose existing single-object datasets into 222k multi-object sequences, totaling 275 hours, with up to 5 objects per sequence. LLM reasoning and geometric checks ensure the compositions remain plausible. On this data, we train a single diffusion transformer for interactions with 1 to 5 objects, predicting each object relative to its interacting body part and generalizing to unseen objects and combinations. Data and code will be made available: 🌐

Capturing every combination of human-object interaction is not feasible. We need compositionally. Most interactions are local. A hand holds a cup, a chair supports the pelvis, while much of the body remains free. COSMI, by Daniel Escandar Ilya Petrov builds on this observation to compose existing single-object datasets into 222k multi-object sequences, totaling 275 hours, with up to 5 objects per sequence. LLM reasoning and geometric checks ensure the compositions remain plausible. On this data, we train a single diffusion transformer for interactions with 1 to 5 objects, predicting each object relative to its interacting body part and generalizing to unseen objects and combinations. Data and code will be made available: 🌐

Gerard Pons-Moll

20,393 views • 2 days ago

For 3D pose some use different keypoints, others SMPL and other models. It's a mess! With Neural Localizer Fields, we can choose the output at test time! allowing to train using any. Results are real time and SOTA across the board. István Sárándi

For 3D pose some use different keypoints, others SMPL and other models. It's a mess! With Neural Localizer Fields, we can choose the output at test time! allowing to train using any. Results are real time and SOTA across the board. István Sárándi

Gerard Pons-Moll

43,598 views • 2 years ago

Check out PhysHead: Simulation-Ready Gaussian Head Avatars (CVPR 2026)! We introduce a layered head & hair model using strands + Gaussian splats, enabling physics-based animation from multi-view video. PS: We didn’t dare compute our bald versions… maybe you’re braver 🙂

Check out PhysHead: Simulation-Ready Gaussian Head Avatars (CVPR 2026)! We introduce a layered head & hair model using strands + Gaussian splats, enabling physics-based animation from multi-view video. PS: We didn’t dare compute our bald versions… maybe you’re braver 🙂

Gerard Pons-Moll

10,232 views • 5 months ago

Nice demo from Unitree. I shaked hands with it but didn't dare to box against it :-)

Nice demo from Unitree. I shaked hands with it but didn't dare to box against it :-)

Gerard Pons-Moll

13,979 views • 1 year ago

No more content to load