Hugging Face Papers ↗ · 2026-11-19
One Prior, Two Dynamics: Repurposing Video Diffusion Transformers for Self-Supervised Latent Action Learning
Media reports Hugging Face Papers Source language: Chinese
O2LaF repurposes a pretrained video diffusion model for self‑supervised latent action learning, extracting both inverse and forward dynamics from unlabeled videos. It achieves state‑of‑the‑art results on few‑shot action probing and zero‑shot action transfer.
Full text
O2LaF repurposes a shared pretrained video model to learn inverse and forward latent action dynamics from unlabeled videos, achieving state-of-the-art few-shot action probing and zero-shot action transfer.