NVIDIA Axolotl3D uses images, masks and point clouds to complete 3D shapes
jonLorraine9 · x · 2026-07-29
NVIDIA’s Spatial Intelligence Lab presents Axolotl3D, a unified framework for faithful 3D shape completion and editing from partial observations.
- The model jointly conditions on images, visibility masks, camera parameters, and partial point clouds.
- The point cloud acts as a geometric anchor, while camera parameters keep multi-view alignment consistent.
- A unified training strategy synthesizes diverse conditioning regimes from large-scale 3D data.
- Experiments on Toys4K and OmniObject3D show state-of-the-art results in clean and occluded settings, plus strong performance in real-world reconstruction and geometry-consistent editing.
Related event: NVIDIA Introduces Axolotl3D for Unified 3D Completion(2 posts)→
More from Research
- Expert Explains AI Verification Crisis: Math is Checkable, Cryptanalysis is Not — matthew_d_green · 2026-07-29
- PrunaVAED claims a faster drop-in decoder for LTX-2.3 video generation — fruesome · 2026-07-29
- Project trains neural networks on Apple Neural Engine via reverse-engineered APIs — maderix · 2026-07-29
- How Puzzles Expose the Reasoning Limits of Current AI — whoamisri · 2026-07-29
- New paper claims decoder LLMs are injective and can reconstruct exact input text — s_scardapane · 2026-07-29
- Seed IQ demos direct vision-to-action play in 3D Doom II — Fit_Transition8824 · 2026-07-29