Research map / Detection and segmentation / CNN features
Dynamic R-CNN: Towards High Quality Object Detection via Dynamic Training
Hongkai Zhang, Hong Chang, Bingpeng Ma, Naiyan Wang, Xilin Chen
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 Dynamic Label Assignment (DLA) (ablation) | Faster R-CNN R50-FPN · COCO val | AP: 37.0 -> 38.2 | Improved |
| add Dynamic SmoothL1 Loss (DSL) (ablation) | Faster R-CNN R50-FPN · COCO val | AP: 37.0 -> 38.0 | Improved |
| add DSL on top of DLA (ablation) | Faster R-CNN R50-FPN · COCO val | AP: 38.2 -> 38.9 | Improved |
| add DLA and DSL, AP90 (ablation) | Faster R-CNN R50-FPN · COCO val | AP: 9.7 -> 15.2 | Improved |
| fix SmoothL1 beta at 2.0 instead of 1.0 (ablation) | Faster R-CNN R50-FPN · COCO val | AP: 37.0 -> 35.9 | Hurt |
| fix SmoothL1 beta at 0.5 instead of 1.0 (ablation) | Faster R-CNN R50-FPN · COCO val | AP: 37.0 -> 37.5 | Improved |
| vary K_I in DLA among 64, 75 and 100 (ablation) | Dynamic R-CNN · COCO val | AP: 38.1, 38.2, 37.9 | No clear effect |
| vary update interval C among 20, 100 and 500 (ablation) | Dynamic R-CNN · COCO val | AP: 38.0, 38.2, 38.1 | No clear effect |
| add dynamic training (DLA+DSL) to Mask R-CNN R50 | Mask R-CNN · COCO val | AP: 37.5 -> 39.4 | Improved |
| add dynamic training (DLA+DSL) to Mask R-CNN R101 | Mask R-CNN · COCO val | AP: 39.7 -> 41.8 | Improved |
| add dynamic training to RetinaNet | RetinaNet R50-FPN · COCO val | AP: 35.6 -> 36.3 | Improved |
| add dynamic training to Faster R-CNN on PASCAL VOC | Faster R-CNN R50-FPN · PASCAL VOC2007 test | AP: 47.2 -> 48.7 | Improved |
Later papers that compared with this method
3 later accepted papers report Dynamic R-CNN* as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Construct Effective Geometry Aware Feature Pyramid Network for Multi-Scale Object Detection (AAAI 2022 · better on 1 of 1 shared result)
- Rank & Sort Loss for Object Detection and Instance Segmentation (ICCV 2021 · better on 2 of 2 shared results)
- Rethinking Transformer-Based Set Prediction for Object Detection (ICCV 2021 · better on 1 of 1 shared result)
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.
- Faster R-CNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · better on 3 of 3 shared results
- Cascade R-CNN — Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 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
- Dynamic Sparse R-CNN (CVPR 2022 · similarity 0.91)
- DynamicDet: A Unified Dynamic Architecture for Object Detection (CVPR 2023 · similarity 0.89)
- Dynamic Refinement Network for Oriented and Densely Packed Object Detection (CVPR 2020 · similarity 0.87)
- Dynamic Convolution: Attention Over Convolution Kernels (CVPR 2020 · similarity 0.86)
- Dynamic DETR: End-to-End Object Detection With Dynamic Attention (ICCV 2021 · similarity 0.86)
- Cascade R-CNN: Delving Into High Quality Object Detection (CVPR 2018 · 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.
