Research map / Detection and segmentation / CNN features
Scale-Aware Trident Networks for Object Detection
Yanghao Li, Yuntao Chen, Naiyan Wang, Zhaoxiang Zhang
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 |
|---|---|---|---|
| add multi-branch trident blocks (ablation) | Faster R-CNN ResNet-101 · COCO minival | AP: 37.9 -> 39.0 | Improved |
| add weight sharing among branches (ablation) | TridentNet ResNet-101 · COCO minival | AP: 39.3 -> 40.6 | Improved |
| add scale-aware training (ablation) | TridentNet ResNet-101 · COCO minival | AP: 40.3 -> 40.6 | Improved |
| add scale-aware training to multi-branch without sharing (ablation) | Multi-branch ResNet-101 · COCO minival | AP: 39.0 -> 39.3 | Improved |
| full TridentNet on deformable backbone | Faster R-CNN ResNet-101-Deformable · COCO minival | AP: 39.9 -> 41.8 | Improved |
| increase dilation rate from 1 to 3 (ablation) | Faster R-CNN ResNet-50 C4 · COCO minival | AP small: 0.174 -> 0.162 | Hurt |
| increase dilation rate from 1 to 3 (ablation) | Faster R-CNN ResNet-50 C4 · COCO minival | AP large: 0.464 -> 0.492 | Improved |
| increase branches from 1 to 3 (ablation) | TridentNet ResNet-50 · COCO minival | AP: 33.2 -> 36.6 | Improved |
| increase branches from 3 to 4 (ablation) | TridentNet ResNet-50 · COCO minival | AP: 36.6 -> 36.5 | No clear effect |
| place trident blocks in conv4 instead of conv2 (ablation) | TridentNet ResNet-50 · COCO minival | AP: 34.1 -> 36.6 | Improved |
| widen major branch scale range to all scales (ablation) | TridentNet Fast · COCO minival | AP: 37.8 -> 39.3 | Improved |
| widen all branch ranges to all scales (ablation) | TridentNet Fast · COCO minival | AP: 37.8 -> 40.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Cascade R-CNN — Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018) · better on 1 of 1 shared result
- SNIPER — SNIPER: Efficient Multi-Scale Training (NeurIPS 2018) · 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
- POD: Practical Object Detection With Scale-Sensitive Network (ICCV 2019 · similarity 0.88)
- An Analysis of Scale Invariance in Object Detection SNIP (CVPR 2018 · similarity 0.88)
- ScaleNet - Improve CNNs through Recursively Rescaling Objects (AAAI 2020 · similarity 0.87)
- SAN: Learning Relationship between Convolutional Features for Multi-Scale Object Detection (ECCV 2018 · similarity 0.87)
- Data-Driven Neuron Allocation for Scale Aggregation Networks (CVPR 2019 · similarity 0.87)
- Feature Pyramid Networks for Object Detection (CVPR 2017 · 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.
