Research map / Detection and segmentation / CNN features
CRFace: Confidence Ranker for Model-Agnostic Face Detection Refinement
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 CRFace confidence ranking to HRNet single-scale detection | HRNet · WiderFace validation Hard | AP (%): 91.5 -> 92.7 | Helped |
| add CRFace confidence ranking rather than multi-scale inference | HRNet · WiderFace validation Hard | AP (%): 92.1 -> 92.7 | Helped |
| add CRFace confidence ranking to RetinaFace single-scale | RetinaFace · WiderFace validation Hard | AP (%): 84.3 -> 88.7 | Helped |
| add CRFace confidence ranking to HAMBox single-scale | HAMBox · WiderFace validation Hard | AP (%): 89.1 -> 91.7 | Helped |
| train CRFace using 8K face-detector outputs | HRNet + CRFace · 8K test set | AP (%): +5.5% | Helped |
| replace regression losses with pairwise ranking loss (ablation) | HRNet confidence refinement · WiderFace validation Hard | AP (%): 91.5 -> 92.7 | Helped |
| use Cross Entropy instead of pairwise ranking loss (ablation) | HRNet confidence refinement · WiderFace validation Hard | AP (%): 91.5 -> 89.7 | Hurt |
| use L1 regression instead of pairwise ranking loss (ablation) | HRNet confidence refinement · WiderFace validation Hard | AP (%): 91.5 -> 90.0 | Hurt |
| use L2 regression instead of pairwise ranking loss (ablation) | HRNet confidence refinement · WiderFace validation Hard | AP (%): 91.5 -> 90.4 | Hurt |
| use Smooth L1 regression instead of pairwise ranking loss (ablation) | HRNet confidence refinement · WiderFace validation Hard | AP (%): 91.5 -> 90.7 | Hurt |
| select 10 pairs rather than 1 pair (ablation) | pairwise ranking loss · WiderFace validation Hard | AP (%): 91.9 -> 92.7 | Helped |
| increase pair count from 10 to 100 (ablation) | pairwise ranking loss · WiderFace validation Hard | AP (%): 92.7 -> 91.9 | Hurt |
Closest papers by meaning
- Sample and Computation Redistribution for Efficient Face Detection (ICLR 2022 · similarity 0.87)
- WIDER FACE: A Face Detection Benchmark (CVPR 2016 · similarity 0.86)
- MogFace: Towards a Deeper Appreciation on Face Detection (CVPR 2022 · similarity 0.86)
- ConfDet: Learning Reliable Confidence for MLLM-based Detection (NeurIPS 2026 · similarity 0.85)
- Attention-Driven Cropping for Very High Resolution Facial Landmark Detection (CVPR 2020 · similarity 0.84)
- CLIB-FIQA: Face Image Quality Assessment with Confidence Calibration (CVPR 2024 · similarity 0.83)
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.