Reproduce “WeVisOne-VL: Unified Visual Perception from Objects to Pixels” 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
Tianming Liang, Junyi Pan, Yuxi Li, Canmiao Fu, Zhipeng Huang, Chong Sun, Chen Li, Jing Lyu
WeVisOne-VL integrates language-guided localization with instance-level masks and alpha mattes. It predicts coordinate tokens and follows each box with an <OBJ> marker, which connects the selected instance to a pixel decoder. On the in-house WeRefMat referring-prompt test, 4B records 9.8 MAD versus 22.4 for Qwen3.5-9B plus SAM2-Matting; this comparison does not establish broader superiority.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.