Research map / Detection and segmentation / Weakly supervised detection and segmentation
Contrastive Learning for Label Efficient Semantic Segmentation
Xiangyun Zhao, Raviteja Vemulapalli, Philip Andrew Mansfield, Boqing Gong, Bradley Green, Lior Shapira, Ying Wu
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 within-image contrastive pretraining before CE fine-tuning | DeepLabV3+ · PASCAL VOC 2012 val (2118 images) | mIoU: 39.1 -> 62.7 | Improved |
| add within-image contrastive pretraining | DeepLabV3+ · Cityscapes val (596 images) | mIoU: 63.6 -> 66.9 | Improved |
| use cross-image loss instead of within-image loss (ablation) | contrastive pretraining · Cityscapes val (596 images) | mIoU: 66.9 -> 67.8 | Improved |
| use cross-image loss instead of within-image loss (ablation) | contrastive pretraining · PASCAL VOC 2012 val (2118 images) | mIoU: 62.7 -> 63.5 | Improved |
| use batch loss instead of within-image loss (ablation) | contrastive pretraining · Cityscapes val (596 images) | mIoU: 66.9 -> 66.0 | Hurt |
| use batch loss instead of within-image loss (ablation) | contrastive pretraining · PASCAL VOC 2012 val (2118 images) | mIoU: 62.7 -> 61.7 | Hurt |
| pretrain then fine-tune instead of joint training of both losses (ablation) | contrastive training strategy · Cityscapes val (596 images) | mIoU: 64.6 -> 66.9 | Improved |
| pretrain then fine-tune instead of joint training of both losses (ablation) | contrastive training strategy · PASCAL VOC 2012 val (2118 images) | mIoU: 39.8 -> 62.7 | Improved |
| add color distortions in contrastive pretraining (ablation) | within-image contrastive pretraining · Cityscapes val (596 images) | mIoU: 66.5 -> 67.5 | Improved |
| add color distortions in contrastive pretraining (ablation) | within-image contrastive pretraining · PASCAL VOC 2012 val (2118 images) | mIoU: 62.9 -> 62.7 | No clear effect |
| add contrastive pretraining to OCR | OCR · Cityscapes val (343 images) | mIoU: 57.1 -> 63.4 | Improved |
| add contrastive pretraining to OCR | OCR · Cityscapes val (596 images) | mIoU: 62.2 -> 65.6 | Improved |
Closest papers by meaning
- Looking Beyond Single Images for Contrastive Semantic Segmentation Learning (NeurIPS 2021 · similarity 0.92)
- Multi-Scale and Cross-Scale Contrastive Learning for Semantic Segmentation (ECCV 2022 · similarity 0.92)
- Exploring Cross-Image Pixel Contrast for Semantic Segmentation (ICCV 2021 · similarity 0.91)
- Cross-Image Pixel Contrasting for Semantic Segmentation (TPAMI 2024 · similarity 0.90)
- Region-Aware Contrastive Learning for Semantic Segmentation (ICCV 2021 · similarity 0.90)
- Pixel Contrastive-Consistent Semi-Supervised Semantic Segmentation (ICCV 2021 · similarity 0.90)
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.
