Research map / Detection and segmentation / CNN features
PnP-DETR: Towards Efficient Visual Analysis With Transformers
Tao Wang, Li Yuan, Yunpeng Chen, Jiashi Feng, Shuicheng Yan
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 |
|---|---|---|---|
| infer with poll ratio 0.5 instead of the trained 0.33 (ablation) | PnP-DETR-R50 · COCO val | AP: 41.1 -> 36.1 | Hurt |
| add pool sampler (30 pool samples) (ablation) | PnP-DETR · sampled COCO | AP: 27.3 -> 28.3 | Improved |
| increase pool samples from 30 to 60 (ablation) | PnP-DETR · sampled COCO | AP: 28.3 -> 28.7 | Improved |
| increase pool samples from 60 to 120 (ablation) | PnP-DETR · sampled COCO | AP: 28.7 -> 28.8 | No clear effect |
| apply pool sampler on full feature set instead of after poll (ablation) | PnP-DETR · sampled COCO | AP: 28.7 -> 28.2 | Hurt |
| replace ranking-based poll sampling with random sampling (no pool) (ablation) | PnP-DETR · sampled COCO | AP: 27.3 -> 22.9 | Hurt |
| replace poll sampling with uniform grid sampling or interpolation (no pool) (ablation) | PnP-DETR · sampled COCO | AP: 27.3 -> 25.9 (uniform), 26.1 (interpolation) | Hurt |
| use 2-layer instead of 1-layer scoring network (ablation) | PnP-DETR · sampled COCO | AP: 27.9 -> 28.7 | Improved |
| use 3-layer instead of 2-layer scoring network (ablation) | PnP-DETR · sampled COCO | AP: 28.7 -> 28.8 | No clear effect |
| reduce scoring network hidden units from 256 to 32 (ablation) | PnP-DETR · sampled COCO | AP: 28.7 -> 28.2 | Hurt |
| raise poll ratio from 0.1 to 0.2 (ablation) | PnP-DETR · sampled COCO | AP: 25.2 -> 27.1 | Improved |
| raise poll ratio from 0.5 to 0.7 (ablation) | PnP-DETR · sampled COCO | AP: 29.2 -> 29.1 | No clear effect |
Closest papers by meaning
- Efficient Decoder-Free Object Detection with Transformers (ECCV 2022 · similarity 0.89)
- Deformable DETR: Deformable Transformers for End-to-End Object Detection (ICLR 2021 · similarity 0.89)
- End-to-End Object Detection with Transformers (ECCV 2020 · similarity 0.87)
- DETRs Beat YOLOs on Real-time Object Detection (CVPR 2024 · similarity 0.85)
- KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling (CVPR 2024 · similarity 0.85)
- Hybrid Proposal Refiner: Revisiting DETR Series from the Faster R-CNN Perspective (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.
