Research map / Detection and segmentation / Weakly supervised detection and segmentation
On Regularized Losses for Weakly-supervised CNN Segmentation
Meng Tang, Federico Perazzi, Abdelaziz Djelouah, Ismail Ben Ayed, Christopher Schroers, Yuri Boykov
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 CRF loss on unlabeled pixels to partial cross entropy (ablation) | DeepLab-MSc-largeFOV · PASCAL VOC12 val (scribbles) | mIoU: 56.0 -> 63.1 | Improved |
| add normalized cut loss to partial cross entropy (ablation) | DeepLab-MSc-largeFOV · PASCAL VOC12 val (scribbles) | mIoU: 56.0 -> 60.5 | Improved |
| use CRF loss instead of normalized cut loss | DeepLab-MSc-largeFOV · PASCAL VOC12 val (scribbles) | mIoU: 60.5 -> 63.1 | Improved |
| add normalized cut term to CRF loss (KernelCut loss) (ablation) | DeepLab-MSc-largeFOV+CRF · PASCAL VOC12 val (scribbles) | mIoU: 65.9 -> 66.7 | Improved |
| add normalized cut term to CRF loss (KernelCut loss) (ablation) | DeepLab-ResNet101+CRF · PASCAL VOC12 val (scribbles) | mIoU: 75.0 -> 75.0 | No clear effect |
| use direct CRF loss instead of online CRF proposal generation (SEC*) | DeepLab-MSc-largeFOV · PASCAL VOC12 val (scribbles) | mIoU: 61.3 -> 63.1 | Improved |
| use direct CRF loss instead of one-time GrabCut proposals | DeepLab-MSc-largeFOV · PASCAL VOC12 val (scribbles) | mIoU: 55.5 -> 63.1 | Improved |
| update proposals online each iteration instead of offline alternation (ScribbleSup) | DeepLab-MSc-largeFOV+CRF · PASCAL VOC12 val (scribbles) | mIoU: 63.1 -> 65.4 | Improved |
| add direct CRF loss to seeding and expansion losses (ablation) | DeepLab-largeFOV · PASCAL VOC12 val (image-level tags) | mIoU: 38.4 -> 43.9 | Improved |
| replace constrain-to-boundary loss with direct CRF loss | SEC · PASCAL VOC12 val (image-level tags) | mIoU: 43.7 -> 43.9 | No clear effect |
| replace constrain-to-boundary loss with direct CRF loss | SEC · PASCAL VOC12 val (image-level tags) | training time (s/batch): 1.19 -> 0.98 | Improved |
| add back-propagation through the CRF inference layer (ablation) | SEC · PASCAL VOC12 val (image-level tags) | mIoU: 43.8 -> 43.7 | No clear effect |
Closest papers by meaning
- Regularized Loss for Weakly Supervised Single Class Semantic Segmentation (ECCV 2020 · similarity 0.94)
- Regularized Loss With Hyperparameter Estimation for Weakly Supervised Single Class Segmentation (TPAMI 2024 · similarity 0.93)
- Normalized Cut Loss for Weakly-Supervised CNN Segmentation (CVPR 2018 · similarity 0.92)
- Beyond Gradient Descent for Regularized Segmentation Losses (CVPR 2019 · similarity 0.91)
- Robust Trust Region for Weakly Supervised Segmentation (ICCV 2021 · similarity 0.91)
- Seed, Expand and Constrain: Three Principles for Weakly-Supervised Image Segmentation (ECCV 2016 · 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.
