Research map / Detection and segmentation / CNN features
Efficient Scale-Permuted Backbone with Learned Resource Distribution
Xianzhi Du, Tsung-Yi Lin, Pengchong Jin, Yin Cui Mingxing Tan, Quoc Le, Xiaodan Song
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 |
|---|---|---|---|
| learn resource distribution over blocks (ablation) | SP-EfficientNet-B0 · COCO val2017 RetinaNet | AP: 33.0 -> 33.8 | Improved |
| learn resource distribution (parameters) (ablation) | SP-EfficientNet-B0 · COCO val2017 | parameters: 4.4M -> 3.6M | Improved |
| add space-to-depth resampling (ablation) | Eff-SpineNet-D0 · COCO val2017 RetinaNet | AP: 33.1 -> 33.8 | Improved |
| add weighted block fusion (ablation) | Eff-SpineNet-D0 · COCO val2017 RetinaNet | AP: 32.8 -> 33.1 | Improved |
| add stochastic depth (ablation) | Eff-SpineNet-D2 · COCO val2017 | AP: +0.4 | Improved |
| replace ReLU with Swish (ablation) | Eff-SpineNet-D2 · COCO val2017 | AP: +2.0 | Improved |
| use large-scale multi-scale training with extended schedule (ablation) | Eff-SpineNet-D2 · COCO val2017 | AP: 32.6 -> 40.1 (+7.4) | Improved |
| add squeeze and excitation (ablation) | Eff-SpineNet-D2 · COCO val2017 | AP: 32.2 -> 32.6 (+0.4) | Improved |
| apply the same training strategies to ResNet-50-FPN: multi-scale training (ablation) | ResNet-50-FPN · COCO val2017 | AP: 37.0 -> 40.4 (+3.4) | Improved |
| apply stochastic depth to ResNet-50-FPN (ablation) | ResNet-50-FPN · COCO val2017 | AP: 40.7 -> 42.3 (+1.6) | Improved |
| lengthen training from 72 to 650 epochs (ablation) | Eff-SpineNet-D2 · COCO val2017 | AP: 34.8 -> 42.5 | Improved |
| use scale-permuted topology instead of EfficientNet-FPN | Eff-SpineNet-D0 · COCO val2017 RetinaNet | AP: 33.5 -> 34.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SpineNet-49 — SpineNet: Learning Scale-Permuted Backbone for Recognition and Localization (CVPR 2020) · better on 0 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
- SpineNet: Learning Scale-Permuted Backbone for Recognition and Localization (CVPR 2020 · similarity 0.91)
- Scale-Aware Trident Networks for Object Detection (ICCV 2019 · similarity 0.86)
- BDCN: Bi-Directional Cascade Network for Perceptual Edge Detection (TPAMI 2020 · similarity 0.85)
- ScaleNet - Improve CNNs through Recursively Rescaling Objects (AAAI 2020 · similarity 0.85)
- Enriched Feature Guided Refinement Network for Object Detection (ICCV 2019 · similarity 0.85)
- Bi-Directional Cascade Network for Perceptual Edge Detection (CVPR 2019 · similarity 0.85)
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.
