Research map / Architectures and efficiency / CNN design
Efficient and Deep Person Re-Identification Using Multi-Level Similarity
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add Level 4 similarity to the Level 2 and Level 3 model (ablation) | Multi-level similarity ReID model · CUHK03 detected | top-1 accuracy: 79.45 -> 86.45 | Helped |
| Add contrastive ranking loss on attended local-part descriptors (ablation) | Level 2 and Level 3 ReID model · CUHK03 detected | top-1 accuracy: 75.90 -> 79.45 | Helped |
| Combine Level 2, Level 3, and Level 4 similarities instead of using one level alone (ablation) | Multi-level similarity ReID model · CUHK03 detected | top-1 accuracy: 74.70; 76.55; 79.15 -> 86.45 | Helped |
| Use Level 3 similarity alone rather than Level 2 similarity alone (ablation) | Single-CSN ReID model · CUHK03 detected | top-1 accuracy: 74.70 -> 76.55 | Helped |
| Use Level 4 similarity alone rather than Level 2 and Level 3 similarity together (ablation) | Multi-level similarity ReID model · CUHK03 detected | top-1 accuracy: 79.45 -> 79.15 | Hurt |
| Do not divide feature maps into three horizontal parts before applying STN (ablation) | Multi-level STN ReID model · CUHK03 detected | top-1 accuracy: 86.45 -> 79.40 | Hurt |
| Replace adaptive STN with fixed central cropping (ablation) | Multi-level ReID model with divided feature maps · CUHK03 detected | top-1 accuracy: 86.45 -> 82.00 | Hurt |
| Add STN to the divided, multi-level similarity model rather than using central cropping (ablation) | Multi-level ReID model · CUHK03 detected | top-1 accuracy: 82.00 -> 86.45 | Helped |
| Use multi-level similarity rather than Level 4 similarity alone (ablation) | ReID model with divided parts and STN · CUHK03 detected | top-1 accuracy: 79.15 -> 86.45 | Helped |
| Add Level 4 similarity to Level 2 and Level 3 similarity | Multi-level similarity ReID model · CUHK03 labeled | top-1 accuracy: 80.30 -> 87.50 | Helped |
| Add Level 4 similarity to Level 2 and Level 3 similarity | Multi-level similarity ReID model · CUHK01 | top-1 accuracy: 86.55 -> 88.20 | Helped |
| Add Level 4 similarity to Level 2 and Level 3 similarity | Multi-level similarity ReID model · VIPeR | top-1 accuracy: 48.03 -> 50.10 | Mixed |
Closest papers by meaning
- Deep Neural Networks with Inexact Matching for Person Re-Identification (NeurIPS 2016 · similarity 0.90)
- A Multi-Task Deep Network for Person Re-Identification (AAAI 2017 · similarity 0.90)
- Group Consistent Similarity Learning via Deep CRF for Person Re-Identification (CVPR 2018 · similarity 0.90)
- Re-Identification With Consistent Attentive Siamese Networks (CVPR 2019 · similarity 0.89)
- Backbone Cannot Be Trained at Once: Rolling Back to Pre-Trained Network for Person Re-Identification (AAAI 2019 · similarity 0.89)
- Consistent-Aware Deep Learning for Person Re-Identification in a Camera Network (CVPR 2017 · similarity 0.89)
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.