Reproduce “AttrSeg: Open-Vocabulary Semantic Segmentation via Attribute Decomposition-Aggregation” 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
Chaofan Ma, Yuhuan Yang, Chen Ju, Fei Zhang, Ya Zhang, Yanfeng Wang
AttrSeg introduces an open-vocabulary semantic segmentation framework that leverages attribute decomposition and hierarchical aggregation to address challenges with ambiguous or novel category names. The model demonstrates superior performance on standard benchmarks and robustly segments objects based solely on attributes, validated by a new "Fantastic Beasts" dataset designed for unfamiliar vocabulary.
Key insights
- Decomposing concepts into attributes, inspired by human cognition, provides richer semantic context for understanding novel categories in vision tasks.
- Detailed attribute descriptions can effectively substitute or complement traditional category names, enabling robust segmentation even for ambiguous or previously unseen concepts.
- A multi-stage, hierarchical aggregation of attributes, incorporating cross-modal interactions, is crucial for effectively combining diverse attribute information into a cohesive semantic representation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.