Reproduce “EVA-Cap: Optimizing Audiovisual Video Captioning via Event-Centric Alignment” 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
Jiapeng Shi, Qiuxia Hou, Jiaqi Leng, Junke Wang, Yanhao Zhang, Haonan Lu, Ziyi Ye, Zuxuan Wu
EVA-Cap trains video captioning to represent audio and visual events, speakers, and timing as explicit targets. It uses two-stage reinforcement-learning training for completeness and synchronization, then attribution, with event-focused data filtering. On ChronusAV’s caption-mediated video-to-audio evaluation, it scores 55.4% accuracy versus AVoCaDO’s 40.9%; this finding is limited to that protocol and does not isolate the contribution of individual components.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.