Research map / Label-efficient and robust learning / Semi-supervised learning
Guided Collaborative Training for Pixel-wise Semi-Supervised Learning
Zhanghan Ke, Di Qiu, Kaican Li, Qiong Yan, Rynson W.H. Lau
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 |
|---|---|---|---|
| use GCT instead of supervised-only training | DeepLab-v2 · Pascal VOC 2012 (1/8 labels) | mIOU: 68.38 -> 72.14 | Improved |
| use GCT instead of supervised-only training | DeepLab-v2 · Pascal VOC 2012 (full labels) | mIOU: 75.32 -> 75.73 | No clear effect |
| use GCT instead of supervised-only training | DHDN · SIDD (1/16 labels) | PSNR (dB): 37.52 -> 38.13 | Improved |
| use GCT instead of supervised-only training | portrait matting model · Portrait matting (100L + 7,700U) | PSNR (dB): 25.39 -> 29.38 | Improved |
| double the unlabeled data from 3,850 to 7,700 | GCT · Portrait matting | PSNR (dB): 27.35 -> 29.38 | Improved |
| use GCT instead of supervised-only training | HDRNet · Night image enhancement | PSNR (dB): 18.72 -> 20.14 | Improved |
| add the dual-model consistency constraint L_dc to the flaw-detector constraint (ablation) | GCT · Pascal VOC 2012 (1/8 labels) | mIOU: 70.04 -> 72.14 | Improved |
| add the flaw-detector constraint L_fc to L_dc (ablation) | GCT · Pascal VOC 2012 (1/8 labels) | mIOU: 71.49 -> 72.14 | Improved |
| raise flaw threshold xi from 0.0 to 0.6 (ablation) | GCT · Pascal VOC 2012 (1/8 labels) | mIOU: 70.04 -> 72.43 | Improved |
| add cosine ramp-up of L_dc (eta from 0 to 3 epochs) (ablation) | GCT · Pascal VOC 2012 (1/8 labels) | mIOU: 71.34 -> 72.14 | Improved |
| vary ramp-up epochs eta among 1, 3, 5, 10 (ablation) | GCT · Pascal VOC 2012 (1/8 labels) | mIOU: 72.03, 72.14, 72.06, 71.95 | No clear effect |
| use flaw detector to disable Mean Teacher consistency where teacher is worse (ablation) | Mean Teacher · Pascal VOC 2012 (1/8 labels) | mIOU: 69.81 -> 70.47 | Improved |
Closest papers by meaning
- Semi-Supervised Semantic Segmentation under Label Noise via Diverse Learning Groups (ICCV 2023 · similarity 0.87)
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.86)
- Semi-Supervised Semantic Segmentation With High- and Low-Level Consistency (TPAMI 2019 · similarity 0.86)
- Mutual Learning for SAM Adaptation: A Dual Collaborative Network Framework for Source-Free Domain Transfer (ICML 2025 · similarity 0.85)
- SemiDAViL: Semi-supervised Domain Adaptation with Vision-Language Guidance for Semantic Segmentation (CVPR 2025 · similarity 0.85)
- Revisiting Semi-Supervised Learning in the Era of Foundation Models (NeurIPS 2025 · 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.
