Research map / Detection and segmentation / CNN features
ProgressFace: Scale-Aware Progressive Learning for Face Detection
Jiashu Zhu, Dong Li, Tiantian Han, Lu Tian, Yi Shan
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 progressive large-to-small training (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Hard | AP: 0.851 -> 0.858 | Improved |
| add progressive large-to-small training (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Easy | AP: 0.937 -> 0.945 | Improved |
| add KL-loss uncertainty estimation (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Hard | AP: 85.8% -> 86.3% | Improved |
| add anchor-free enhancement module (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Hard | AP: 86.3% -> 87.9% | Improved |
| train baseline for longer (ablation) | ProgressFace (MobileNet-0.25) baseline · WIDER FACE val Hard | AP: 85.1% -> 85.3% | No clear effect |
| use progressive training at equal epochs (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Hard | AP: 85.3% -> 85.8% | Improved |
| add anchor-free branch to anchor-based training (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Hard | AP: 0.851 -> 0.860 | Improved |
| use anchor-free only instead of anchor-based only (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Hard | AP: 0.851 -> 0.813 | Hurt |
| place anchor-free module on P3 rather than P5 (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Hard | AP: 0.864 -> 0.879 | Improved |
| add anchor-free module to every anchor-based branch (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Hard | AP | No clear effect |
| combine anchor-free outputs at test time (ablation) | ProgressFace (MobileNet-0.25) · WIDER FACE val Hard | AP: 0.879 -> 0.876 | Hurt |
| use ProgressFace-Light instead of RetinaFace + DCNv2 | MobileNet-0.25 · WIDER FACE val Hard | AP: 79.5% -> 87.9% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DSFD — DSFD: Dual Shot Face Detector (CVPR 2019) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Finding Tiny Faces (CVPR 2017 · similarity 0.90)
- MogFace: Towards a Deeper Appreciation on Face Detection (CVPR 2022 · similarity 0.89)
- Sample and Computation Redistribution for Efficient Face Detection (ICLR 2022 · similarity 0.88)
- WIDER FACE: A Face Detection Benchmark (CVPR 2016 · similarity 0.88)
- Group Sampling for Scale Invariant Face Detection (CVPR 2019 · similarity 0.88)
- Scale-Aware Face Detection (CVPR 2017 · similarity 0.88)
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.
