Reproduce “Multi-scale 3D Convolution Network for Video Based Person Re-Identification” 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
Jianing Li, Shiliang Zhang, Tiejun Huang
A two-stream Multi-scale 3D (M3D) convolution network, incorporating a Residual Attention Layer (RAL), was developed to enhance video-based person re-identification by efficiently extracting and fusing spatial and multi-scale temporal features. This architecture achieved state-of-the-art performance on benchmarks like MARS, with 74.06% mAP and 84.39% rank-1 accuracy, while maintaining higher efficiency and compactness compared to existing 3D CNN approaches.
Key insights
- Multi-scale temporal information, efficiently captured by the M3D layer, significantly boosts ReID performance while maintaining model compactness.
- A residual attention mechanism with factorized spatio-temporal masks can effectively refine features by emphasizing discriminative regions and segments in video sequences.
- Combining distinct spatial appearance features and learned temporal motion features through a two-stream approach yields superior and more robust ReID representations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.