Research map / Detection and segmentation / CNN features
NAS-FCOS: Fast Neural Architecture Search for Object Detection
Check 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 both the FPN and prediction head (ablation) | FCOS with ResNet-50 · COCO | AP: 37.4 -> 39.8 | Helped |
| Search the FPN while keeping the original head (ablation) | FCOS with ResNet-50 · COCO | AP: 37.4 -> 38.9 | Helped |
| Search the head while keeping the original FPN (ablation) | FCOS with ResNet-50 · COCO | AP: 37.4 -> 38.7 | Helped |
| Replace the original head with the searched head (ablation) | FCOS with ResNet-50 · COCO | AP: 37.4 -> 38.7 | Helped |
| Replace standard 3x3 FPN convolutions with deformable 3x3 convolutions (ablation) | Original FCOS FPN · COCO | AP: 37.4 -> 38.4 | Helped |
| Use the searched FPN and head instead of the deformable-convolution FCOS FPN (ablation) | FCOS with ResNet-50 · COCO | AP: 38.4 -> 39.8 | Helped |
| Use validation loss sum rather than AP as the architecture-search reward (ablation) | NAS-FCOS reinforcement-learning search · VOC proxy task | AP | Helped |
| Use searched NAS-FCOS instead of FCOS with the same backbone | MobileNetV2 detector · COCO test-dev | AP: 31.2 -> 32.0 | Helped |
| Use searched NAS-FCOS instead of FCOS with the same backbone | ResNet-50 detector · COCO test-dev | AP: 37.4 -> 39.8 | Helped |
| Use searched NAS-FCOS instead of FCOS with the same backbone | ResNet-101 detector · COCO test-dev | AP: 41.5 -> 43.0 | Helped |
| Use searched NAS-FCOS instead of FCOS with the same backbone | ResNeXt-101 64x4d detector · COCO test-dev | AP: 43.2 -> 44.5 | Helped |
| Use NAS-FCOS with improvements instead of improved FCOS | ResNeXt-101 64x4d detector · COCO test-dev | AP: 44.7 -> 46.1 | Helped |
Closest papers by meaning
- SM-NAS: Structural-to-Modular Neural Architecture Search for Object Detection (AAAI 2020 · similarity 0.93)
- NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection (CVPR 2019 · similarity 0.92)
- SP-NAS: Serial-to-Parallel Backbone Search for Object Detection (CVPR 2020 · similarity 0.92)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.91)
- FBNetV3: Joint Architecture-Recipe Search Using Predictor Pretraining (CVPR 2021 · similarity 0.90)
- Joint-DetNAS: Upgrade Your Detector With NAS, Pruning and Dynamic Distillation (CVPR 2021 · 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.