Research map / Detection and segmentation / CNN features
Finding Tiny Faces
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 fixed large-scale context to small-face detection (ablation) | Fixed-size face detector · WIDER FACE small faces | accuracy: 18.9% | Helped |
| Add context beyond the face extent (ablation) | Fixed-size face detector · WIDER FACE large faces | accuracy: 1.5% | Helped |
| Extend tiny-face context beyond 300x300 pixels (ablation) | Tiny-face detector · WIDER FACE tiny faces | accuracy | Hurt |
| Remove high-resolution foveal features, using only res4 (ablation) | Small-face template · WIDER FACE small faces | performance: 7% | Hurt |
| Remove high-resolution foveal features, using only res5 (ablation) | Small-face template · WIDER FACE small faces | performance: 33% | Hurt |
| Remove foveal structure from the template (ablation) | Large-face template · WIDER FACE large faces | performance | No clear effect |
| Use a medium-size template on 2X upsampled test images for small faces (ablation) | Fixed-size face detector · WIDER FACE small faces (25x20) | performance: 69% to 75% | Helped |
| Use a medium-size template on images downsampled by 2X for large faces (ablation) | Fixed-size face detector · WIDER FACE large faces (250x200) | performance: 89% to 94% | Helped |
| Use 2X template resolution for small faces (ablation) | Scale-specific detector · WIDER FACE small faces (25x20) | accuracy: 6.3% | Helped |
| Use 0.5X template resolution for large faces (ablation) | Scale-specific detector · WIDER FACE large faces (250x200) | accuracy: 5.6% | Helped |
| Prune redundant templates (ablation) | HR-ResNet101 · WIDER FACE validation | AP: 0.919 -> 0.925 (Easy); 0.908 -> 0.914 (Medium); 0.823 -> 0.831 (Hard) | Helped |
| Remove hard-example mining (ablation) | HR-ResNet101 · WIDER FACE validation | AP: 0.919 -> 0.917 (Easy); 0.908 -> 0.904 (Medium); 0.823 -> 0.825 (Hard) | No clear effect |
Closest papers by meaning
- Group Sampling for Scale Invariant Face Detection (CVPR 2019 · similarity 0.89)
- MogFace: Towards a Deeper Appreciation on Face Detection (CVPR 2022 · similarity 0.88)
- WIDER FACE: A Face Detection Benchmark (CVPR 2016 · similarity 0.88)
- Scale-Aware Face Detection (CVPR 2017 · similarity 0.88)
- An Analysis of Scale Invariance in Object Detection SNIP (CVPR 2018 · similarity 0.88)
- ScaleKD: Distilling Scale-Aware Knowledge in Small Object Detector (CVPR 2023 · 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.