Research map / Detection and segmentation / CNN features
Dynamic Sparse R-CNN
Check 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 |
|---|---|---|---|
| add dynamic proposal generation with the staircase structure (ablation) | Sparse R-CNN · COCO 2017 val | AP: 45.0 -> 45.7 | Helped |
| add the staircase feature aggregation to DPG (ablation) | DPG · COCO 2017 val | AP: 45.3 -> 45.7 | Helped |
| add dynamic many-to-one assignment using OTA with dynamic q=8 (ablation) | Sparse R-CNN with DPG · COCO 2017 val | AP: 45.7 -> 46.0 | Helped |
| add the unit increasing strategy (ablation) | Sparse R-CNN with DPG and DLA · COCO 2017 val | AP: 46.0 -> 47.2 | Helped |
| replace Hungarian matching with OTA using fixed k=1 (ablation) | Sparse R-CNN · COCO 2017 val | AP: 45.0 -> 44.7 | Hurt |
| replace Hungarian matching with OTA using fixed k=2 (ablation) | Sparse R-CNN · COCO 2017 val | AP: 45.0 -> 45.9 | Helped |
| use dynamic k estimation with q=8 instead of fixed k (ablation) | Sparse R-CNN with OTA · COCO 2017 val | AP: 45.0 -> 46.1 | Helped |
| add two losses for OTA matching (ablation) | Sparse R-CNN with dynamic q=8 · COCO 2017 val | AP: 46.1 -> 46.1 | No clear effect |
| vary dynamic k estimation from q=4 to q=9 (ablation) | Dynamic Sparse R-CNN · COCO 2017 val | AP: q=4: 46.7; q=5: 46.7; q=6: 46.7; q=7: 46.4; q=8: 47.2; q=9: 46.1 | Mixed |
| vary the number of proposal experts from 3 to 5 (ablation) | DPG · COCO 2017 val | AP: 3: 45.4; 4: 45.7; 5: 45.3 | Mixed |
| add DPG and DLA | Sparse R-CNN with ResNet-50 · COCO 2017 val | AP: 45.0 -> 47.2 | Helped |
Closest papers by meaning
- Sparse R-CNN: End-to-End Object Detection With Learnable Proposals (CVPR 2021 · similarity 0.93)
- Group R-CNN for Weakly Semi-Supervised Object Detection With Points (CVPR 2022 · similarity 0.88)
- G-CNN: An Iterative Grid Based Object Detector (CVPR 2016 · similarity 0.86)
- Designing BERT for Convolutional Networks: Sparse and Hierarchical Masked Modeling (ICLR 2023 · similarity 0.86)
- Hybrid Proposal Refiner: Revisiting DETR Series from the Faster R-CNN Perspective (CVPR 2024 · similarity 0.86)
- SparseFormer: Sparse Visual Recognition via Limited Latent Tokens (ICLR 2024 · similarity 0.84)
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.