Reproduce “Beyond Token Representations: Explicit Visual Object Grounding for Video Reasoning 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
Jun Huang, Junyang Zhang, Zijie Yue, Bingkun Wang, Yong Luo, Miaojing Shi
EVOG performs reasoning video segmentation by asking a multimodal model to generate a bounding box and seven points for a queried object, refining point prompts through dialogue and using them to guide mask decoding and propagation. On ReVOS’s reasoning split, its 7B version scores 63.3% J&F, versus UniPixel’s 61.5%; its full system runs at 1.32 FPS, versus 2.48 for the baseline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.