Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised Segmentation based on Error-Correcting Supervision
Robert Mendel, Luis Antonio de Souza Jr, David Rauber, João Paulo Papa, Christoph Palm
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 Error-Correcting Supervision with 1/8 labels | DeepLabv3+ ResNet50 · Pascal VOC 2012 val (1/8 labeled) | mIoU: 65.20 -> 70.22 | Improved |
| add Error-Correcting Supervision with 1/8 labels | DeepLabv3+ ResNet50 · Cityscapes val (1/8 labeled) | mIoU: 63.12 -> 67.38 | Improved |
| add Error-Correcting Supervision with 1/4 labels | DeepLabv3+ ResNet50 · Pascal VOC 2012 val (1/4 labeled) | mIoU: 69.78 -> 72.60 | Improved |
| add Error-Correcting Supervision with 1/4 labels | DeepLabv3+ ResNet50 · Cityscapes val (1/4 labeled) | mIoU: 67.31 -> 70.70 | Improved |
| add Error-Correcting Supervision with 1/2 labels | DeepLabv3+ ResNet50 · Pascal VOC 2012 val (1/2 labeled) | mIoU: 73.42 -> 74.63 | Improved |
| add Error-Correcting Supervision with 1/2 labels | DeepLabv3+ ResNet50 · Cityscapes val (1/2 labeled) | mIoU: 71.22 -> 72.89 | Improved |
| use Negated Focal Loss instead of Cross-Entropy on pseudo-labels (ablation) | ECS lambda_cor 0.1 · Pascal VOC 2012 val (1/4 labeled) | mIoU: 68.39 -> 72.60 | Improved |
| use Negated Focal Loss instead of Cross-Entropy on pseudo-labels (ablation) | ECS lambda_cor 0.1 · Cityscapes val (1/4 labeled) | mIoU: 69.27 -> 70.70 | Improved |
| add auxiliary corrector regularization (lambda_cor 0.1) to pseudo-label training (ablation) | ECS with NFL · Pascal VOC 2012 val (1/4 labeled) | mIoU: 71.97 -> 72.60 | Improved |
| add auxiliary regularization alone, without semi-supervised learning (ablation) | supervised DeepLabv3+ · Pascal VOC 2012 val (1/4 labeled) | mIoU: 69.79 -> 69.93 | No clear effect |
| add auxiliary regularization alone, without semi-supervised learning (ablation) | supervised DeepLabv3+ · Cityscapes val (1/4 labeled) | mIoU: 67.31 -> 68.22 | Improved |
| use a DeepLabv3+ discriminator with the adversarial method of Hung et al. | adversarial semi-supervised baseline · Pascal VOC 2012 val (1/4 labeled) | mIoU: 69.79 -> 65.83 | Hurt |
Closest papers by meaning
- Semi-Supervised Semantic Segmentation With Error Localization Network (CVPR 2022 · similarity 0.92)
- Semi-Supervised Semantic Segmentation under Label Noise via Diverse Learning Groups (ICCV 2023 · similarity 0.91)
- Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels (CVPR 2022 · similarity 0.90)
- Semi-Supervised Semantic Segmentation With High- and Low-Level Consistency (TPAMI 2019 · similarity 0.90)
- Semi-Supervised Semantic Image Segmentation With Self-Correcting Networks (CVPR 2020 · similarity 0.90)
- RWMS: Reliable Weighted Multi-Phase for Semi-supervised Segmentation (AAAI 2024 · similarity 0.88)
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.
