Reproduce “V-JEPA-LM: Video-Language Models That Read Dynamics Beyond Static, Text-Aligned Representations” 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
Keuntae Kim, Dongjun Kim, Yong Suk Choi
The authors train a video-language model with questions answered by transformations of public clips, alongside captions and question-answer data. In a matched 2B experiment, counterfactual supervision raises V-JEPA’s twelve-score average by 4.5 points at 8k steps. Qwen3-4B V-JEPA-LM scores 72.14 versus 61.08 for the best open 4B baseline; this aggregate does not mean it wins every benchmark. The study targets clip-level dynamics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.