Research map / Detection and segmentation / CNN features
Auto-FPN: Automatic Network Architecture Adaptation for Object Detection Beyond Classification
Hang Xu, Lewei Yao, Wei Zhang, Xiaodan Liang, Zhenguo Li
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 Auto-FPN architecture (neck and head) instead of FPN | FPN · COCO minval | mAP: 38.6 -> 40.5 | Improved |
| search Auto-FPN architecture instead of FPN | FPN · PASCAL VOC | mAP: 79.7 -> 81.8 | Improved |
| search Auto-FPN architecture instead of FPN | FPN · BDD | mAP: 33.0 -> 33.9 | Improved |
| use Auto-FPN with ResNet-101 instead of FPN | FPN ResNet-101 · COCO | mAP: 40.7 -> 42.5 | Improved |
| transfer COCO-searched architecture to Visual Genome | FPN · Visual Genome | mAP: 7.0 -> 7.2 | Improved |
| transfer COCO-searched architecture to ADE | FPN · ADE | mAP: 10.5 -> 12.1 | Improved |
| apply Auto-fusion to SSD512 | SSD512 · COCO | mAP: 29.3 -> 31.8 | Improved |
| apply Auto-fusion to SSD512 | SSD512 · PASCAL VOC | mAP: 79.8 -> 79.9 | No clear effect |
| add step training (optimize architecture parameters in the middle of search) (ablation) | Auto-fusion · PASCAL VOC (search) | mAP: 81.7 -> 82.2 | Improved |
| freeze the backbone during search (ablation) | Auto-fusion · PASCAL VOC (search) | mAP: 65.2 -> 82.2 | Improved |
| search without batch norm (versus with BN) (ablation) | Auto-head · PASCAL VOC (search) | mAP: 78.3 -> 81.3 | Improved |
| load pretrained neck when searching head (ablation) | Auto-head · PASCAL VOC (search) | mAP: 78.4 -> 81.3 | Improved |
Later papers that compared with this method
2 later accepted papers report Auto-FPN as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Hit-Detector: Hierarchical Trinity Architecture Search for Object Detection (CVPR 2020 · better on 1 of 1 shared result)
- SP-NAS: Serial-to-Parallel Backbone Search for Object Detection (CVPR 2020 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SM-NAS: Structural-to-Modular Neural Architecture Search for Object Detection (AAAI 2020 · similarity 0.92)
- SP-NAS: Serial-to-Parallel Backbone Search for Object Detection (CVPR 2020 · similarity 0.91)
- Fast Neural Network Adaptation via Parameter Remapping and Architecture Search (ICLR 2020 · similarity 0.91)
- NAS-FCOS: Fast Neural Architecture Search for Object Detection (CVPR 2020 · similarity 0.91)
- NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection (CVPR 2019 · similarity 0.90)
- FNA++: Fast Network Adaptation via Parameter Remapping and Architecture Search (TPAMI 2020 · similarity 0.90)
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.
