StreamMAE Makes Self-Supervised Learning Work on Continuous Video Streams

A NeurIPS paper introduces StreamMAE, enabling self-supervised pretraining on 95 hours of continuous video. The authors found high intra-batch similarity—not inter-batch—is the real obstacle, and fixing it brings performance close to iid sampling.

2026-10-09 ~ 2026-10-09 · 4 related posts

• StreamMAE: self-supervised pretraining on 95h of continuous video matches i.i.d. training — y_m_asano · 2026-10-09
• NeurIPS paper explains why streaming SSL fails and how to fix it — y_m_asano · 2026-10-09
• Fixing intra-batch similarity nearly restores iid performance in streaming learning — y_m_asano · 2026-10-09

1 near-duplicate retellings: y_m_asano