The authors formulate open-ended detailed video captioning as an active perception problem.
OACap turns video captioning into active, tool-guided observation
A caption-guided agent iteratively identifies missing or unsupported evidence, calls spatiotemporal tools to acquire it, and improves downstream video question answering across multiple backbones.
Chinese Tech
Hao Shi · Hong Yan · Xiaoyang Zhang · Qing Cui · Chao Huang
Ant Group · Zhejiang University · Peking University
Research Digest··2 min read
Shi et al.
§