Research map / Detection and segmentation / CNN features
SM-NAS: Structural-to-Modular Neural Architecture Search for Object Detection
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 |
|---|---|---|---|
| Search the detector structure and modules jointly, then refine module architectures | Object detectors · COCO test-dev | AP | Helped |
| Use GN instead of BN for scratch training (ablation) | FPN ResNet-50 · COCO val | mAP: 24.8 -> 29.4 | Helped |
| Add WS to GN for scratch training (ablation) | FPN ResNet-50 · COCO val | mAP: 29.4 -> 30.7 | Helped |
| Increase learning rate from the GN+WS scratch setting (ablation) | FPN ResNet-50 · COCO val | mAP: 30.7 -> 36.4 | Helped |
| Use larger learning rate and batch size with GN+WS and train for 16 epochs (ablation) | FPN ResNet-50 · COCO val | mAP: 30.7 -> 37.5 | Helped |
| Train from scratch with BN rather than ImageNet pretraining (ablation) | FPN ResNet-50 · COCO val | mAP: 36.5 -> 24.8 | Hurt |
| Increase BN scratch-training batch size and learning rate (ablation) | FPN ResNet-50 · COCO val | mAP: 24.8 -> 28.3 | Helped |
| Transfer the searched E0 architecture rather than use FPN with ResNet-50 | Object detection architecture · VOC 2007 test | mAP: 80.4 -> 81.4 | Helped |
| Use the searched E3 architecture rather than FPN with ResNet-101 | Object detection architecture · BDD | mAP: 36.9 -> 39.6 | Helped |
Closest papers by meaning
- SP-NAS: Serial-to-Parallel Backbone Search for Object Detection (CVPR 2020 · similarity 0.94)
- NAS-FCOS: Fast Neural Architecture Search for Object Detection (CVPR 2020 · similarity 0.93)
- Hit-Detector: Hierarchical Trinity Architecture Search for Object Detection (CVPR 2020 · similarity 0.92)
- DetNAS: Backbone Search for Object Detection (NeurIPS 2019 · similarity 0.91)
- Joint-DetNAS: Upgrade Your Detector With NAS, Pruning and Dynamic Distillation (CVPR 2021 · similarity 0.90)
- G-NAS: Generalizable Neural Architecture Search for Single Domain Generalization Object Detection (AAAI 2024 · similarity 0.88)
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.