Reproduce “THG-Omni: Traceable Human-Centric Graph Supervision for Video Reasoning” 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
Jisheng Dang, Kai Wei, Ronghao Lin, Bimei Wang, Hong Peng, Wei-Shi Zheng, Bin Hu, Qi Tian and 1 more
THG-Omni organizes tracked people and time-bound claims into graphs, then trains an omni-modal model to use graph fields for video questions. On IntentBench, it scores 71.72% versus 55.27% for the unadapted Qwen2.5-Omni-7B backbone. On NExT-QA, matched graphs yield 89.54% accuracy versus 85.10% without graph input for the same trained model. Evaluation graphs remain automatically predicted and unverified.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.