Research map / Detection and segmentation / CNN features
RecursiveDet: End-to-End Region-Based Recursive Object Detection
Jing Zhao, Li Sun, Qingli Li
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 |
|---|---|---|---|
| share decoder parameters across cascade stages (ablation) | Sparse R-CNN · COCO val2017 (R50) | AP: 42.8 -> 42.4 AP, 106M -> 40M params | Mixed |
| reuse dynamic layer and out layer within a stage (ablation) | Sparse R-CNN with shared stages · COCO val2017 (R50) | AP: 42.4 -> 43.6 | Improved |
| keep first decoder stage independent of recursive stages (ablation) | RecursiveDet · COCO val2017 (R50) | AP: 43.6 -> 43.9 | Improved |
| add bounding box positional encoding to the recursive decoder (ablation) | RecursiveDet · COCO val2017 (R50) | AP: 43.9 -> 44.4 | Improved |
| add bounding box PE to non-recursive Sparse R-CNN (ablation) | Sparse R-CNN · COCO val2017 (R50) | AP: 42.8 -> 43.1 | Improved |
| add centerness-based PE (ablation) | RecursiveDet · COCO val2017 (R50) | AP: 44.4 -> 45.2 | Improved |
| make centerness mask learnable instead of static (ablation) | RecursiveDet · COCO val2017 (R50) | AP: 45.2 -> 44.9 | Hurt |
| predict center offsets to adjust centerness peak (ablation) | RecursiveDet · COCO val2017 (R50) | AP: 45.2 -> 44.5 | Hurt |
| apply RecursiveDet to Sparse R-CNN with ResNet-50 | Sparse R-CNN · COCO val2017 | AP: 45.0 -> 46.5 AP, 106M -> 55M params | Improved |
| apply RecursiveDet to AdaMixer with ResNet-50 | AdaMixer · COCO val2017 | AP: 47.0 -> 47.9 AP, 135M -> 43M params | Improved |
| apply RecursiveDet to DiffusionDet with ResNet-101 | DiffusionDet · COCO val2017 | AP: 46.6 -> 46.9 AP, 130M -> 76M params | Improved |
| add test-time augmentation | RecursiveDet · COCO test-dev (Swin-B) | AP: 53.1 -> 55.1 | 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.
- Sparse R-CNN — Sparse R-CNN: End-to-End Object Detection With Learnable Proposals (CVPR 2021) · better on 5 of 5 shared results
- AdaMixer — AdaMixer: A Fast-Converging Query-Based Object Detector (CVPR 2022) · better on 4 of 4 shared results
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
- R-FCN: Object Detection via Region-based Fully Convolutional Networks (NeurIPS 2016 · similarity 0.88)
- Object Detection Based on Region Decomposition and Assembly (AAAI 2019 · similarity 0.87)
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016 · similarity 0.87)
- Efficient Decoder-Free Object Detection with Transformers (ECCV 2022 · similarity 0.87)
- Hybrid Proposal Refiner: Revisiting DETR Series from the Faster R-CNN Perspective (CVPR 2024 · similarity 0.85)
- Theoretically Achieving Continuous Representation of Oriented Bounding Boxes (CVPR 2024 · 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.
