Research map / Detection and segmentation / CNN features
Sample and Computation Redistribution for Efficient Face Detection
Read the paperCheck 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 Sample Redistribution by extending crop scales from [0.3, 1.0] to [0.3, 2.0] (ablation) | ResNet-2.5GF · WIDER FACE validation, Hard | AP: 67.32 -> 74.47 | Helped |
| Apply computation redistribution to the backbone only (ablation) | ResNet-2.5GF · WIDER FACE validation, Hard | AP: 67.32 -> 69.78 | Helped |
| Apply computation redistribution across backbone, neck, and head (ablation) | ResNet-2.5GF · WIDER FACE validation, Hard | AP: 67.32 -> 70.98 | Helped |
| Use two-step rather than one-step whole-detector computation redistribution (ablation) | ResNet-2.5GF · WIDER FACE validation, Hard | AP: 70.98 -> 71.37; 0.39% | Helped |
| Combine Sample Redistribution with two-step Computation Redistribution (ablation) | ResNet-2.5GF · WIDER FACE validation, Hard | AP: 71.37 -> 77.87 | Helped |
| Redistribute computation to shallow backbone stages and optimize the detector structure | SCRFD-34GF vs ResNet-34GF · WIDER FACE validation, Hard | AP: 84.02 -> 85.29 | Helped |
| Redistribute computation to shallow backbone stages and optimize the detector structure | SCRFD-10GF vs ResNet-10GF · WIDER FACE validation, Hard | AP: 80.42 -> 83.05 | Helped |
| Redistribute computation to shallow backbone stages and optimize the detector structure | SCRFD-2.5GF vs ResNet-2.5GF · WIDER FACE validation, Hard | AP: 74.47 -> 77.87 | Helped |
| Redistribute computation for the mobile regime | SCRFD-0.5GF vs MobileNet-0.5GF · WIDER FACE validation, Hard | AP: 66.68 -> 68.51 | Helped |
| Use VGA-resolution rather than large-scale testing | TinaFace · WIDER FACE validation, Hard | AP: 0.930 -> 0.814 | Hurt |
Closest papers by meaning
- MogFace: Towards a Deeper Appreciation on Face Detection (CVPR 2022 · similarity 0.89)
- CRFace: Confidence Ranker for Model-Agnostic Face Detection Refinement (CVPR 2021 · similarity 0.87)
- Scale-Aware Face Detection (CVPR 2017 · similarity 0.86)
- WIDER FACE: A Face Detection Benchmark (CVPR 2016 · similarity 0.86)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.84)
- From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes (NeurIPS 2025 · similarity 0.84)
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.