Research map / Detection and segmentation / CNN features
Detecting Faces Using Inside Cascaded Contextual CNN
Kaipeng Zhang, Zhanpeng Zhang, Hao Wang, Zhifeng Li, Yu Qiao, Wei 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 body part sensitive learning (BPSL) (ablation) | R-Net-2 baseline · Validation set (R-Net-2 face/non-face classification) | overall accuracy: 95.92% -> 96.67% | Improved |
| add inside cascaded structure (ICS) (ablation) | R-Net-2 baseline · Validation set (R-Net-2 face/non-face classification) | overall accuracy: 95.92% -> 97.12% | Improved |
| add both ICS and BPSL (ablation) | R-Net-2 baseline · Validation set (R-Net-2 face/non-face classification) | overall accuracy: 95.92% -> 97.43% | Improved |
| add ICS (ablation) | R-Net-2 baseline · Validation set (R-Net-2 face/non-face classification) | positives accuracy: 90.78% -> 92.18% | Improved |
| add BPSL (ablation) | R-Net-2 baseline · Validation set (R-Net-2 face/non-face classification) | negatives accuracy: 97.63% -> 98.48% | Improved |
| add ICS and BPSL (ablation) | R-Net-2 baseline · Validation set (R-Net-2 face/non-face classification) | negatives accuracy: 97.63% -> 99.06% | Improved |
| set DR layer threshold theta to 0.01 instead of 0 (deeply supervised net) (ablation) | R-Net-2 (lambda 0) · Validation set (R-Net-2 face/non-face classification) | classification accuracy | Improved |
| increase body part loss weight lambda from 0 up to 0.04 (ablation) | R-Net-2 without ICS · Validation set (R-Net-2 face/non-face classification) | classification accuracy | Mixed |
| add BPSL for faces with variant body poses and without body region (ablation) | two-stream CNN · FDDB subsets | FDDB subset of 400 + 400 faces | Improved |
| add ICS and BPSL to the full framework (ablation) | R-Net-1 and R-Net-2 cascade · FDDB | FDDB ROC | Improved |
| use proposed method versus state-of-the-art on WIDER FACE hard set | full framework · WIDER FACE | WIDER FACE precision-recall | Improved |
Closest papers by meaning
- Supervised Transformer Network for Efficient Face Detection (ECCV 2016 · similarity 0.89)
- Joint Training of Cascaded CNN for Face Detection (CVPR 2016 · similarity 0.89)
- Faceness-Net: Face Detection through Deep Facial Part Responses (TPAMI 2017 · similarity 0.88)
- PyramidBox: A Context-assisted Single Shot Face Detector (ECCV 2018 · similarity 0.87)
- Finding Tiny Faces (CVPR 2017 · similarity 0.87)
- Unconstrained Still/Video-Based Face Verification with Deep Convolutional Neural Networks (IJCV 2017 · similarity 0.86)
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.
