Building Training Supervision Offline Using 40k Videos
mangahomanga · x · 2026-07-18
The author details the sources of their training data and supervision signals:
- Offline processing of 40,000 Something-Something-V2 interaction videos to extract object masks, metric geometry, camera motion, and dense 3D tracks.
- Future frames are used strictly as supervision targets, not as model inputs.
- This means the model learns to predict future motion based solely on past observations, without relying on future information leakage.
Related event: MotionForesight predicts future 3D motion from existing video models(6 posts)→
More from Research
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- Terence Tao's SAIR Foundation revamps Mathathon with $1.5M compute per team — _sathvikr · 2026-10-27
- Mathathon 2.0 launches with Terence Tao's SAIR, $1.5M in compute for teams — _sathvikr · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27
- PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction — kwangmoo_yi · 2026-10-27
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27