Reproduce “OnlineRefer: A Simple Online Baseline for Referring Video Object Segmentation” 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
Dongming Wu, Tiancai Wang, Yuang Zhang, Xiangyu Zhang, Jianbing Shen
OnlineRefer presents an online framework for referring video object segmentation that processes videos frame-by-frame, utilizing explicit cross-frame query propagation. This method achieves state-of-the-art performance on major benchmarks like Refer-Youtube-VOS and Refer-DAVIS17, surpassing previous clip-based offline approaches.
Key insights
- A well-designed online model with explicit query propagation can achieve and even surpass the performance of complex offline RVOS methods.
- Propagating the content query and using the previous frame's predicted bounding box as a refined spatial prior are critical for effective and continuous temporal association.
- A simple, heuristic-free approach for query propagation, by selecting the single highest-scoring query, is more effective than more complex query management strategies.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.