Research map / Video / Video segmentation and tracking
Appearance-Preserving 3D Convolution for Video-based Person Re-identification
Xinqian Gu, Hong Chang, Bingpeng Ma, Hongkai Zhang, Xilin Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace I3D 3D convolution with AP3D (add APM) | I3D, ResNet-50 · MARS | top-1 / mAP: 88.6 -> 90.1 top-1; 83.0 -> 84.8 mAP | Improved |
| replace P3D-C 3D convolution with AP3D | P3D-C, ResNet-50 · MARS | top-1 / mAP: 88.5 -> 90.1 top-1; 83.1 -> 85.1 mAP | Improved |
| replace P3D-B 3D convolution with AP3D | P3D-B, ResNet-50 · DukeMTMC-VideoReID | mAP: 95.3 -> 95.9 | Improved |
| use plain I3D or P3D instead of 2D baseline | C2D, ResNet-50 · MARS | top-1: 88.9 -> 88.6 (I3D) | No clear effect |
| add Contrastive Attention to AP-P3D (ablation) | AP-P3D · MARS | mAP: 84.8 -> 85.1 | Improved |
| add Contrastive Attention to AP-I3D (ablation) | AP-I3D · MARS | top-1: 89.7 -> 90.1 | Improved |
| place one AP3D block in stage 1 instead of none (ablation) | C2D, ResNet-50 · MARS | top-1 / mAP: 88.9 -> 89.0 top-1; 83.4 -> 83.2 mAP | No clear effect |
| place one AP3D block in stage 4 (ablation) | C2D, ResNet-50 · MARS | top-1 / mAP: 88.9 -> 88.8 top-1; 83.4 -> 82.9 mAP | Hurt |
| place one AP3D block in stage 3 (ablation) | C2D, ResNet-50 · MARS | top-1 / mAP: 88.9 -> 89.7 top-1; 83.4 -> 84.1 mAP | Improved |
| increase AP3D blocks from 5 to 10 (ablation) | AP3D stage 2,3 · MARS | mAP: 85.1 -> 84.7 | Hurt |
| increase AP3D blocks from 2 to 5 (ablation) | AP3D stage 2,3 · MARS | mAP: 84.7 -> 85.1 | Improved |
| replace Non-local with AP3D | C2D, ResNet-50 · MARS | mAP: 85.0 -> 85.1 | No clear effect |
Closest papers by meaning
- Multi-Scale 3D Convolution Network for Video Based Person Re-Identification (AAAI 2019 · similarity 0.86)
- Learning 3D Shape Feature for Texture-Insensitive Person Re-Identification (CVPR 2021 · similarity 0.86)
- SSN3D: Self-Separated Network to Align Parts for 3D Convolution in Video Person Re-Identification (AAAI 2021 · similarity 0.85)
- Exploring the Camera Bias of Person Re-identification (ICLR 2025 · similarity 0.84)
- From Poses to Identity: Training-Free Person Re-Identification via Feature Centralization (CVPR 2025 · similarity 0.84)
- Distilling CLIP with Dual Guidance for Learning Discriminative Human Body Shape Representation (CVPR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
