Research map / Architectures and efficiency / Attention and transformers
Effectiveness of Vision Transformer for Fast and Accurate Single-Stage Pedestrian Detection
Jing Yuan, Barmpoutis Panagiotis, Tania Stathaki
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 enhancement module (ablation) | CSP · Caltech test | Reasonable MR-2: 6.8 -> 3.7 | Improved |
| add enhancement module (ablation) | CSP · Caltech test | Heavy MR-2: 50.7 -> 42.8 | Improved |
| add enhancement module (ablation) | CSP · Citypersons val | Heavy MR-2: 41.8 -> 38.6 | Improved |
| add enhancement module (ablation) | CSP · Citypersons val | Reasonable MR-2: 11.7 -> 10.9 | Improved |
| add enhancement module on three levels (ablation) | SSD300 · Caltech test, IoU 0.5 | Reasonable MR-2: 28.6 -> 22.6 | Improved |
| add enhancement module on three levels (ablation) | Progressive refinement detector · Caltech test | Reasonable MR-2: 13.4 -> 12.3 | Improved |
| add enhancement module (inference cost) (ablation) | CSP · Caltech test | FPS: 39.2 -> 29.5 | Hurt |
| add enhancement module (parameters) (ablation) | CSP · Model size | total parameters: 39.96 M -> 32.24 M | Improved |
| use 2 encoders instead of 1 (ablation) | CSP + deformable-encoder feature enhancement · Caltech test | Reasonable MR-2: 3.7 -> 4.9 | Hurt |
| use 4 encoders instead of 1 (ablation) | CSP + deformable-encoder feature enhancement · Caltech test | Heavy MR-2: 42.8 -> 45.1 | Hurt |
| input scales 1/4,1/8,1/16 instead of 1/4,1/8,1/8 (ablation) | CSP + deformable-encoder feature enhancement · Caltech test | Reasonable MR-2: 5.9 -> 4.1 | Improved |
| concatenate enhanced maps with L2Norm instead of add or separate (ablation) | CSP + deformable-encoder feature enhancement · Caltech test | Reasonable and Heavy MR-2: 3.7 and 42.8 best | Improved |
Closest papers by meaning
- Generalizable Pedestrian Detection: The Elephant in the Room (CVPR 2021 · similarity 0.85)
- Deformable DETR: Deformable Transformers for End-to-End Object Detection (ICLR 2021 · similarity 0.85)
- Learning Efficient Single-stage Pedestrian Detectors by Asymptotic Localization Fitting (ECCV 2018 · similarity 0.85)
- Embracing Single Stride 3D Object Detector With Sparse Transformer (CVPR 2022 · similarity 0.85)
- Improving Occlusion and Hard Negative Handling for Single-Stage Pedestrian Detectors (CVPR 2018 · similarity 0.84)
- Efficient Decoder-Free Object Detection with Transformers (ECCV 2022 · 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.
