Reproduce “STAIR: Spatial-Temporal Reasoning with Auditable Intermediate Results for Video Question Answering” 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
Yueqian Wang, Yuxuan Wang, Kai Chen, Dongyan Zhao
STAIR breaks video questions into smaller tasks that localize actions, identify objects, and apply temporal relations such as “after.” Its modules return task-specific outputs that can be inspected, and some are trained with intermediate supervision from scene-graph-derived answers. On AGQA 2.0, STAIR reaches 52.06% overall accuracy versus MAC’s 49.67% using the same I3D features, while supervision lowers accuracy to 51.41%.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.