Research map / Detection and segmentation / CNN features
Faceness-Net: Face Detection through Deep Facial Part Responses
Shuo Yang, Ping Luo, Chen Change Loy, Xiaoou Tang
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 Faceness re-scoring to EdgeBox proposals | EdgeBox · FDDB face proposals | proposals needed for 90% detection rate: 600 -> 288 | Improved |
| add Faceness re-scoring to MCG proposals | MCG · FDDB face proposals | proposals needed for 90% detection rate: 942 -> 158 | Improved |
| add Faceness re-scoring to Selective Search proposals | Selective Search · FDDB face proposals | proposals needed for 90% detection rate: 641 -> 237 | Improved |
| add Faceness re-scoring of template proposals (ablation) | Template Proposal · FDDB face proposals | proposals needed for 90% detection rate: 847 -> 351 | Improved |
| share representation across part networks (Faceness-SR) instead of five separate networks | Faceness · FDDB face proposals (MCG) | proposals needed for 75% detection rate: 13 -> 69 | Hurt |
| share representation across part networks (Faceness-SR) at high recall | Faceness · FDDB face proposals (MCG) | proposals needed for 90% detection rate: 158 -> 158 | No clear effect |
| train on uncropped instead of cropped images (ablation) | attribute-aware networks · LFW part detection (350 proposals) | part detection rate: Eye 95.87% -> 97.19%; Hair 95.56% -> 94.57% | No clear effect |
| fine-tune with part-level attributes instead of face/non-face or all 25 attributes (ablation) | attribute-aware networks · FDDB face proposals | detection rate | Improved |
| pretrain on facial attributes instead of ImageNet (ablation) | multi-task CNN · face detection fine-tuning | iterations to converge: 45,000 vs more than 200,000 iterations | Improved |
| use template proposal (TP) instead of generic proposals | Faceness-Net · FDDB | detection rate at high recall: over 96.5% | Improved |
| use template proposal instead of generic proposals at fewer than 300 proposals | Faceness-Net · FDDB | detection rate | Hurt |
| add sharing representation and template proposal (SR-TP) | Faceness-Net · FDDB discrete score | recall: 90.99% -> +1.85% | Improved |
Closest papers by meaning
- Detecting Faces Using Inside Cascaded Contextual CNN (ICCV 2017 · similarity 0.88)
- Attributes for Improved Attributes: A Multi-Task Network Utilizing Implicit and Explicit Relationships for Facial Attribute Classification (AAAI 2017 · similarity 0.87)
- Finding Tiny Faces (CVPR 2017 · similarity 0.86)
- HyperFace: A Deep Multi-Task Learning Framework for Face Detection, Landmark Localization, Pose Estimation, and Gender Recognition (TPAMI 2017 · similarity 0.86)
- EAC-Net: Deep Nets with Enhancing and Cropping for Facial Action Unit Detection (TPAMI 2018 · similarity 0.85)
- Scale-Aware Face Detection (CVPR 2017 · similarity 0.85)
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.
