ProMo3D: motion cues are decodable from frozen 3D foundation model tokens with 160 videos
zhenjun_zhao · x · 2026-10-30
NeurIPS-accepted ProMo3D shows that static-dynamic separation is decodable from frozen feed-forward 3D foundation model tokens (π³X, VGGT, DA3), concentrated in a low-dimensional subspace. Its Probing-based Motion Distillation (PMD) turns pairwise motion-compatibility scores into dense motion masks, trained on just 160 YouTube-VOS videos on a single 16 GB GPU. It transfers zero-shot to DAVIS, SegTrackv2 and FBMS-59, with π³X-PMD+SAM2 hitting 81.9/92.3 J-M/J-R on DAVIS17 — rivaling far heavier supervised methods like GeoMotion while being lightweight and data-efficient.
More from Research
- RLNF: Training AI Models on Human Preferences Directly from Brain Signals — alex_verem · 2026-10-30
- Impagliazzo, Paturi and Williams launch site to scrutinize OpenAI's claimed ETH refutation — fortnow · 2026-10-30
- Know, then Draw: UCSD & Adobe Cut Image-Gen Training Compute 48x While Boosting DPG-Bench 10% — Yucheng Mao · 2026-10-30
- Live mechinterp session hunts 'neurons of sadness' in Qwen3.5 — burny_tech · 2026-10-30
- RT-SAFE Benchmark Shows Faster Reasoning Doesn't Mean Safer Agents in Real-Time Embodied Environments — Lianhuiq · 2026-10-30
- LeCun team's LeHami: symplectic Hamiltonian flow keeps JEPA world models from collapsing latent states — ylecun · 2026-10-30