Research map / Detection and segmentation / Weakly supervised detection and segmentation
Regularized Loss for Weakly Supervised Single Class Semantic Segmentation
Olga Veksler
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 |
|---|---|---|---|
| start from OxfordPet-pretrained weights instead of annealing (ablation) | UMobV2 regularized loss · ECSSD | F-measure: 57.96 -> 79.20 | Improved |
| start from OxfordPet-pretrained weights instead of annealing (ablation) | UResNext regularized loss · ECSSD | F-measure: 73.47 -> 83.49 | Improved |
| start from OxfordPet-pretrained weights instead of annealing (ablation) | UResNext regularized loss · DUT-OMRON | F-measure: 64.48 -> 73.42 | Improved |
| start from OxfordPet-pretrained weights instead of annealing (ablation) | UMobV2 regularized loss · MSRA-B | F-measure: 86.20 -> 85.51 | Hurt |
| start from OxfordPet-pretrained weights instead of annealing (ablation) | UMobV2 regularized loss · DUT-OMRON | F-measure: 53.79 -> 54.49 | No clear effect |
| add annealing of sigma and complete loss (all parts) (ablation) | regularized loss training · single-class datasets | training success | Improved |
| use larger UResNext backbone instead of UMobV2 | regularized loss · DUT-OMRON | F-measure: 53.79 -> 64.48 | Improved |
| train with ground truth instead of regularized loss | UMobV2 · ECSSD | F-measure: 57.96 -> 84.78 | Improved |
| add negative loss using other categories as negatives (ablation) | UnetX · Pascal VOC cosegmentation | Jaccard (mean): 0.59 -> 0.62 | Improved |
| train on larger Web dataset instead of the small cosegmentation set | UnetX with negative loss · Pascal VOC cosegmentation | Jaccard (mean): 0.62 -> 0.67 | Improved |
| add negative loss, train on PascalSingle (ablation) | DeepLab-ResNet101 pseudo-labels · Pascal VOC 2012 val | mIoU: 66.7 -> 67.1 | Improved |
| train on PascalSingle instead of Web images | DeepLab-ResNet101 pseudo-labels · Pascal VOC 2012 val | mIoU: 64.0 -> 66.7 | Improved |
Closest papers by meaning
- Regularized Loss With Hyperparameter Estimation for Weakly Supervised Single Class Segmentation (TPAMI 2024 · similarity 0.96)
- On Regularized Losses for Weakly-supervised CNN Segmentation (ECCV 2018 · similarity 0.94)
- Seed, Expand and Constrain: Three Principles for Weakly-Supervised Image Segmentation (ECCV 2016 · similarity 0.93)
- Normalized Cut Loss for Weakly-Supervised CNN Segmentation (CVPR 2018 · similarity 0.92)
- Combining Bottom-Up, Top-Down, and Smoothness Cues for Weakly Supervised Image Segmentation (CVPR 2017 · similarity 0.92)
- Webly Supervised Semantic Segmentation (CVPR 2017 · 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.
