Research map / Architectures and efficiency / Neural architecture search
Auto-ReID: Searching for a Part-Aware ConvNet for Person Re-Identification
Ruijie Quan, Xuanyi Dong, Yu Wu, Linchao Zhu, Yi Yang
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 |
|---|---|---|---|
| search directly on the reID dataset instead of CIFAR-10 (DARTS baseline vs DARTS network) (ablation) | DARTS · Market-1501 (no ImageNet pretraining) | mAP (%): 65.2 -> 68.5 | Improved |
| add the part-aware module to the search space (ablation) | DARTS baseline search · Market-1501 (no ImageNet pretraining) | mAP (%): 68.5 -> 74.6 | Improved |
| replace classification search loss with retrieval loss (ablation) | Auto-ReID on reID search space · Market-1501 (no ImageNet pretraining) | mAP (%): 71.3 -> 72.7 | Mixed |
| add ImageNet-pretrained search with the searched part-aware CNN versus the best baseline CNN (ablation) | Auto-ReID · Market-1501 (with ImageNet pretraining) | mAP (%): 83.4 -> 85.1 | Improved |
| add part-aware module via search instead of PCB attached at tail (ablation) | baseline searched CNN · Market-1501 (no ImageNet pretraining) | mAP (%): 70.2 -> 74.6 | Improved |
| attach PCB part module at tail of GDAS network (ablation) | GDAS · Market-1501 (no ImageNet pretraining) | mAP (%): 66.8 -> 68.1 | Mixed |
| increase training channels C from 16 to 32 (ablation) | searched CNN (search C=32, l=[2,2,2,2]) · Market-1501 (no ImageNet pretraining) | mAP (%): 58.3 -> 68.3 | Improved |
| train with deeper l=[3,4,6,3] when searched with l=[2,2,2,2] (ablation) | searched CNN (C=32 training) · Market-1501 (no ImageNet pretraining) | mAP (%): 68.3 -> 64.9 | Hurt |
| add re-ranking | Auto-ReID · Market-1501 (with ImageNet pretraining) | mAP (%): 85.1 -> 94.2 | Improved |
| search architecture on MSMT17 instead of Market-1501 with default hyper-parameters (ablation) | Auto-ReID · MSMT17 | accuracy | Hurt |
Closest papers by meaning
- A Multi-Task Deep Network for Person Re-Identification (AAAI 2017 · similarity 0.91)
- Backbone Cannot Be Trained at Once: Rolling Back to Pre-Trained Network for Person Re-Identification (AAAI 2019 · similarity 0.90)
- Interaction-And-Aggregation Network for Person Re-Identification (CVPR 2019 · similarity 0.89)
- Combined Depth Space Based Architecture Search for Person Re-Identification (CVPR 2021 · similarity 0.89)
- Multi-Prompts Learning with Cross-Modal Alignment for Attribute-Based Person Re-identification (AAAI 2024 · similarity 0.87)
- Instruct-ReID: A Multi-purpose Person Re-identification Task with Instructions (CVPR 2024 · similarity 0.87)
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.
