Reproduce “Mutual-Supervised Feature Modulation Network for Occluded Pedestrian 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
Ye He, Chao Zhu, Xu-Cheng Yin
The Mutual-Supervised Feature Modulation Network (MSFMN) improves occluded pedestrian detection by introducing feature-level mutual supervision between full and visible body representations. This approach, from the University of Science and Technology Beijing, achieved a 9.43% absolute reduction in log-average miss rate on the CityPersons Heavy Occlusion subset and outperformed prior state-of-the-art methods on both CityPersons and Caltech datasets.
Key insights
- Feature-level mutual supervision via MSFMM encourages the full body detector to learn more robust and complete features by integrating contextual information from occluding parts, guided by visible body features.
- Configuring the visible body branch solely for classification improves overall performance, preventing degradation from conflicting feature requirements for regression.
- Focusing the MSFMM similarity loss calculation exclusively on positive training samples yields more effective feature modulation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.