Research map / Detection and segmentation / CNN features
Group Sampling for Scale Invariant Face Detection
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 |
|---|---|---|---|
| Balance positive and negative training samples separately for each anchor scale (ablation) | FPN-finest · WIDER FACE validation All | AP: 80.2 -> 82.8 | Helped |
| Use one P2 feature layer rather than multiple FPN prediction layers, keeping anchor strides comparable (ablation) | FPN · WIDER FACE validation All | AP: 82.1 vs. 81.6 | No clear effect |
| Add group sampling to the P2 detector (ablation) | P2 face detector · WIDER FACE validation All | AP: 80.3 -> 82.8 | Helped |
| Add group sampling to the P3 detector (ablation) | P3 face detector · WIDER FACE validation All | AP: 79.6 -> 83.7 | Helped |
| Add group sampling to the P4 detector (ablation) | P4 face detector · WIDER FACE validation All | AP: 74.4 -> 79.6 | Helped |
| Replace Smooth-L1 regression with least-square IoU loss (ablation) | FPN-finest-sampling · WIDER FACE validation All | AP: 82.8 -> 83.7 | Helped |
| Use −ln(IoU) regression loss instead of Smooth-L1 (ablation) | FPN-finest-sampling · WIDER FACE validation All | AP: 82.8 -> 83.5 | Helped |
| Add Fast R-CNN to the group-sampled detector (ablation) | FPN-finest-sampling · WIDER FACE validation All | AP: 82.7 -> 83.9 | Helped |
| Add group sampling to the Fast R-CNN stage (ablation) | FPN-finest-sampling + Fast R-CNN + IoU loss · WIDER FACE validation All | AP: 84.6 -> 85.7 | Helped |
| Use group sampling rather than OHEM or Focal Loss (ablation) | FPN-finest · WIDER FACE validation All | AP: 81.5 -> 82.8; 81.2 -> 82.8 | Helped |
| Increase the number of sampled training examples per scale (ablation) | FPN-finest-sampling · WIDER FACE validation All | AP | Helped |
Closest papers by meaning
- Finding Tiny Faces (CVPR 2017 · similarity 0.89)
- GroupFace: Learning Latent Groups and Constructing Group-Based Representations for Face Recognition (CVPR 2020 · similarity 0.87)
- Scale-Aware Face Detection (CVPR 2017 · similarity 0.87)
- MogFace: Towards a Deeper Appreciation on Face Detection (CVPR 2022 · similarity 0.87)
- An Analysis of Scale Invariance in Object Detection SNIP (CVPR 2018 · similarity 0.86)
- Sample and Computation Redistribution for Efficient Face Detection (ICLR 2022 · 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.