Reproduce “VCD: A Dataset for Visual Commonsense Discovery in Images” 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

Xiangqing Shen, Fanfan Wang, Siwei Wu, Rui Xia

The authors introduce VCD, a dataset that grounds visible and inferred commonsense triples to objects in images. It combines Visual Genome annotations with ConceptNet relations across 11 categories, covering 106,277 images and approximately 14 million object–commonsense pairs. On OK-VQA, adding generated commonsense raises Qwen-VL-7B’s reported score from 58.6 to 60.0, a result limited to this evaluation setup.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.