Reproduce “From Objects to Events: Unlocking Complex Visual Understanding in Object Detectors via LLM-guided Symbolic 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
Yuhui Zeng, Haoxiang Wu, Wenjie Nie, Xiawu Zheng, Guangyao Chen, Yunhang Shen, Jun Peng, Yonghong Tian and 1 more
The SymbolicDet framework augments standard object detectors with complex event understanding through LLM-guided symbolic reasoning, achieving significant AUROC improvements for diverse event detection tasks in a training-free and interpretable manner.
Key insights
- Standard object detectors implicitly contain rich visual information that can be explicitly interpreted for complex event understanding when combined with symbolic reasoning.
- Large Language Models can effectively guide symbolic search by infusing world knowledge and semantic understanding in their native text domain, separating perception from reasoning for enhanced interpretability.
- The synergistic combination of evolutionary algorithms for pattern discovery and LLM guidance enables the training-free, interpretable recognition of complex events from existing visual detections.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.