Research map / Detection and segmentation / CNN features
Cascade R-CNN: Delving Into High Quality 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 |
|---|---|---|---|
| replace a single FPN+ detector with a three-stage Cascade R-CNN | FPN+ ResNet-101 · COCO test-dev | AP: 38.8 -> 42.8 | Helped |
| replace a single FPN+ detector with a three-stage Cascade R-CNN | FPN+ ResNet-50 · COCO val | AP: 36.5 -> 40.3 | Helped |
| replace a single R-FCN detector with a three-stage Cascade R-CNN | R-FCN ResNet-50 · COCO val | AP: 27.0 -> 31.1 | Helped |
| replace a single R-FCN detector with a three-stage Cascade R-CNN | R-FCN ResNet-101 · COCO val | AP: 30.3 -> 33.3 | Helped |
| replace a single Faster R-CNN detector with a three-stage Cascade R-CNN | Faster R-CNN VGG · COCO val | AP: 23.6 -> 27.0 | Helped |
| replace a single FPN+ detector with a three-stage Cascade R-CNN | FPN+ · COCO val | AP: 34.9 -> 38.9 | Helped |
| replace the baseline with iterative application of one detector | FPN+ · COCO val | AP: 34.9 -> 35.4 | Helped |
| replace the baseline with integral-loss classifiers | FPN+ · COCO val | AP: 34.9 -> 35.4 | Helped |
| add specialized cascade stages instead of relying on one detector (ablation) | FPN+ · COCO val | AP: 34.9 -> 38.3 | Helped |
| increase IoU thresholds across stages rather than using u=0.5 throughout (ablation) | Cascade R-CNN · COCO val | AP: 36.8 -> 38.5 | Helped |
| add sequential regression statistics (ablation) | Cascade R-CNN with increasing IoU thresholds · COCO val | AP: 38.5 -> 38.9 | Helped |
| add a fourth detection stage at u=0.75 (ablation) | three-stage Cascade R-CNN · COCO val | AP: 38.9 -> 38.6 | Hurt |
Closest papers by meaning
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.91)
- Sparse R-CNN: End-to-End Object Detection With Learnable Proposals (CVPR 2021 · similarity 0.91)
- CRAFT Objects From Images (CVPR 2016 · similarity 0.90)
- RON: Reverse Connection With Objectness Prior Networks for Object Detection (CVPR 2017 · similarity 0.89)
- R-FCN++: Towards Accurate Region-Based Fully Convolutional Networks for Object Detection (AAAI 2018 · similarity 0.89)
- Feature Pyramid Networks for Object Detection (CVPR 2017 · 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.