HuC-VideoMAE: Human-Centric Video Masked Autoencoding from synthetic data

Key points

• Recent high-quality synthetic video datasets generated from motion-capture data, such as BEDLAM2.0, offer a promising ethical alternative.1
• In this work, we investigate self-supervised pretraining of video transformers on synthetic human-motion datasets.1
• To address this limitation, we propose a human-centric masking scheme that leverages body keypoints and person bounding box regions.1
• To promote the use of ethical action recognition models, we will publicly release our pretrained models.1

Sources (1)

• [1]HuC-VideoMAE: Human-Centric Video Masked Autoencoding from synthetic data
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 02:30 PM

  Modern action recognition models rely on video transformers pretrained on massive collections of web-crawled videos, such as Kinetics-700.

  Recent high-quality synthetic video datasets generated from motion-capture data, such as BEDLAM2.0, offer a promising ethical alternative.

Extractive summary: sentences quoted from the sources.