Reproduce “FastInstShadow: A Simple Query-Based Model for Instance Shadow Detection” 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
Takeru Inoue, Ryusuke Miyamoto
Meiji University researchers developed FastInstShadow (FIS), the first query-based model for instance shadow detection that directly learns object-shadow relationships within a single framework. The model achieves state-of-the-art accuracy on the SOBA dataset while operating at real-time speeds of over 30 frames per second.
Key insights
- Integrating the association mechanism directly into a query-based detection framework allows for learning intrinsic relationships between objects and shadows.
- A sequential dual-path transformer decoder can effectively process object and shadow features to learn their mutual dependencies within a unified model.
- Specific training strategies like shadow direction learning and box-aware mask loss improve the accuracy and spatial precision of shadow and object mask predictions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.