Research map / Detection and segmentation / CNN features
Efficient Neural Architecture Transformation Search in Channel-Level for Object Detection
Junran Peng, Ming Sun, ZHAO-XIANG ZHANG, Tieniu Tan, Junjie Yan
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 |
|---|---|---|---|
| Apply NATS to the ResNet-50 backbone | Faster R-CNN · COCO minival | AP: 36.4 -> 38.4 | Improved |
| Apply NATS to the ResNet-50 backbone | Mask R-CNN · COCO minival | AP: 37.5 -> 39.3 | Improved |
| Apply NATS to the ResNet-50 backbone | Cascade R-CNN · COCO minival | AP: 40.7 -> 42.0 | Improved |
| Apply NATS to the ResNet-50 backbone | RetinaNet · COCO minival | AP: 36.0 -> 37.3 | Improved |
| Apply NATS with 32 channels per group | ResNet-101 Faster R-CNN · COCO minival | AP: 38.6 -> 40.4 | Improved |
| Apply NATS with 32 channels per group | ResNeXt-101 32×4d Faster R-CNN · COCO minival | AP: 40.5 -> 41.6 | Improved |
| Increase the number of channel groups from 1 to 16 (ablation) | NATS on ResNet-50 · COCO minival | AP: 36.9 -> 38.4 | Improved |
| Increase the number of channel groups from 16 to 32 (ablation) | NATS on ResNet-50 · COCO minival | AP: 38.4 -> 38.2 | Hurt |
| Set channel-group width to 32 rather than 64 channels (ablation) | NATS on ResNet-50 · COCO minival | AP: 37.8 -> 38.3 | Improved |
| Use denser dilation candidates instead of sparse candidates (ablation) | NATS on ResNet-50 · COCO minival | AP: 37.7 -> 38.0 | Improved |
| Add aspect-ratio dilation candidates to the dense dilation set (ablation) | NATS on ResNet-50 · COCO minival | AP: 38.0 -> 38.4 | Improved |
Closest papers by meaning
- Neural Architecture Search: A Survey (JMLR 2019 · similarity 0.88)
- Hit-Detector: Hierarchical Trinity Architecture Search for Object Detection (CVPR 2020 · similarity 0.87)
- Auto-FPN: Automatic Network Architecture Adaptation for Object Detection Beyond Classification (ICCV 2019 · similarity 0.87)
- Can Weight Sharing Outperform Random Architecture Search? An Investigation With TuNAS (CVPR 2020 · similarity 0.86)
- Hierarchical Representations for Efficient Architecture Search (ICLR 2018 · similarity 0.86)
- Are Labels Necessary for Neural Architecture Search? (ECCV 2020 · similarity 0.86)
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.
