Research map / Detection and segmentation / CNN features
PyramidBox: A Context-assisted Single Shot Face Detector
Xu Tang, Daniel K. Du, Zeqiang He, Jingtuo Liu
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 LFPN (top-down from conv_fc7) to the S3FD baseline (ablation) | S3FD baseline · WIDER FACE val Hard | mAP: +1.9% | Improved |
| start the feature pyramid from the top layer conv7_2 (plain FPN) instead of conv_fc7 (ablation) | LFPN · WIDER FACE val Hard | mAP: 85.9 -> 86.1 (LFPN) | Hurt |
| start the pyramid too low (conv4_3) (ablation) | LFPN · WIDER FACE val Hard | mAP: 86.1 -> 84.8 | Hurt |
| add data-anchor-sampling on top of LFPN (ablation) | PyramidBox · WIDER FACE val Easy/Medium/Hard | mAP: +0.4%, +0.4% and +0.6% | Improved |
| add PyramidAnchors (K=2, s_pa=2.0) to baseline (ablation) | S3FD baseline · WIDER FACE val Easy/Medium/Hard | mAP: +0.7%, +0.6% and +0.9% | Improved |
| use one pyramid level K=1 with s_pa=1.5 (ablation) | PyramidAnchors · WIDER FACE val Easy | mAP: 94.0 -> 93.8 | No clear effect |
| increase anchor levels from K=1 to K=2 (s_pa=2.0) (ablation) | PyramidAnchors · WIDER FACE val Hard | mAP: 84.9 -> 85.1 | Improved |
| replace DSSD prediction module with CPM (ablation) | prediction module · WIDER FACE val Hard | mAP: 88.2 -> 88.5 | Improved |
| replace SSH context module with CPM (ablation) | prediction module · WIDER FACE val Hard | mAP: 88.4 -> 88.5 | Improved |
| add max-in-out layer (ablation) | PyramidBox · WIDER FACE val Easy/Medium/Hard | mAP: +0.2%, +0.3% and +0.1% | Improved |
| add context-sensitive prediction module after pyramid anchors (ablation) | PyramidBox · WIDER FACE val Hard | mAP: 88.3 -> 88.8 | Improved |
| combine all components (LFPN, DAS, PA, CPM, max-in-out) (ablation) | S3FD baseline · WIDER FACE val Easy/Medium/Hard | mAP: +2.1%, +2.3% and +4.7% | Improved |
Closest papers by meaning
- MogFace: Towards a Deeper Appreciation on Face Detection (CVPR 2022 · similarity 0.88)
- Detecting Faces Using Inside Cascaded Contextual CNN (ICCV 2017 · similarity 0.87)
- Transformer-Based Context Condensation for Boosting Feature Pyramids in Object Detection (IJCV 2023 · similarity 0.87)
- Finding Tiny Faces (CVPR 2017 · similarity 0.87)
- ProgressFace: Scale-Aware Progressive Learning for Face Detection (ECCV 2020 · similarity 0.87)
- BFBox: Searching Face-Appropriate Backbone and Feature Pyramid Network for Face Detector (CVPR 2020 · similarity 0.87)
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.
