Reproduce “Memory-augmented Dense Predictive Coding for Video Representation Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tengda Han, Weidi Xie, Andrew Zisserman
MemDPC learns action-relevant video features without action labels during pretraining. It uses learned memory bank entries to form weighted mixtures of candidate future features, rather than predicting one future frame. On UCF101, a controlled R18 RGB comparison at 128×128 reports 68.2% top-1 accuracy versus 63.6% for reimplemented DPC after downstream fine-tuning; this result is limited to that setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.