Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised Skin Detection by Network With Mutual Guidance
Yi He, Jiayuan Shi, Chuan Wang, Haibin Huang, Jiaming Liu, Guanbin Li, Risheng Liu, Jue Wang
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 mutual guidance between skin and body decoders (UNet) (ablation) | dual-task network · Own skin validation set | IoU (%): 76.56 -> 80.11 | Improved |
| add mutual guidance (DeepLab-v3-MobileNet backbone) (ablation) | dual-task network · Own skin validation set | IoU (%): 75.15 -> 79.02 | Improved |
| add CRF loss to cross-entropy (ablation) | semi-supervised loss · Own skin validation set | IoU (%): 79.28 -> 79.48 | Improved |
| add WCE loss to cross-entropy (ablation) | semi-supervised loss · Own skin validation set | IoU (%): 79.28 -> 79.52 | Improved |
| add both CRF and WCE losses to cross-entropy (ablation) | semi-supervised loss · Own skin validation set | IoU (%): 79.28 -> 81.18 | Improved |
| reduce skin training data from 5k to 1k, keeping 5k body | Ours · Own skin validation set | IoU (%): 81.18 -> 75.29 | Hurt |
| use informative stage-1 guidance (body box, GMM skin) instead of zeros (ablation) | dual-task network (UNet) · Own skin validation set | IoU (%): 80.11 -> 80.74 | Improved |
| add gradient stopping from stage-2 guidance to stage-1 decoders (ablation) | dual-task network · qualitative skin examples | mask quality | Improved |
| finetune stage 2 after stage-1 training rather than training from scratch (ablation) | dual-task network · Own skin validation set | IoU | Improved |
| use shared encoder with extra body data, even without mutual guidance (ablation) | dual-task network · Own skin validation set | IoU (%): 76.74 vs 75.59 | Improved |
| use smaller DeepLab-v3-MobileNet backbone instead of UNet (about 60% of parameters) (ablation) | dual-task network · Own skin validation set | IoU (%): 80.11 -> 79.02 | Mixed |
Closest papers by meaning
- BPJDet: Extended Object Representation for Generic Body-Part Joint Detection (TPAMI 2024 · similarity 0.83)
- Tasks Integrated Networks: Joint Detection and Retrieval for Image Search (TPAMI 2020 · similarity 0.83)
- Relational Learning for Joint Head and Human Detection (AAAI 2020 · similarity 0.83)
- PairDETR : Joint Detection and Association of Human Bodies and Faces (CVPR 2024 · similarity 0.82)
- Detecting Hands and Recognizing Physical Contact in the Wild (NeurIPS 2020 · similarity 0.81)
- Distribution Matching for Multi-Task Learning of Classification Tasks: A Large-Scale Study on Faces & Beyond (AAAI 2024 · similarity 0.80)
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.
