Research map / Detection and segmentation / Weakly supervised detection and segmentation
Robust Trust Region for Weakly Supervised Segmentation
Dmitrii Marin, 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 |
|---|---|---|---|
| replace gradient descent on relaxed Potts loss with discrete trust region (alpha-expansion) | Grid-GD · ScribbleSup val, full scribbles | mIoU: 0.64 -> 0.67 | Improved |
| replace gradient descent on relaxed Potts loss with trust region | Grid-GD · ScribbleSup val, clicks (length 0) | mIoU: 0.54 -> 0.57 | Improved |
| replace gradient descent on relaxed Potts loss with trust region | Grid-GD · ScribbleSup val, length 0.5 | mIoU: 0.62 -> 0.64 | Improved |
| use trust region on 8-grid Potts instead of dense CRF gradient descent | Dense-GD · ScribbleSup val, full scribbles | mIoU: 0.64 -> 0.67 | Improved |
| add Potts regularized loss via trust region to partial cross-entropy | PCE-GD · ScribbleSup val, full scribbles | mIoU: 0.61 -> 0.67 | Improved |
| add Potts regularized loss via trust region to partial cross-entropy | PCE-GD · ScribbleSup val, clicks (length 0) | mIoU: 0.50 -> 0.57 | Improved |
| add Potts regularization with plain gradient descent (Grid-GD) to partial cross-entropy | PCE-GD · ScribbleSup val, full scribbles | mIoU: 0.61 -> 0.64 | Improved |
| use a fixed set of proposals (lambda=0) instead of lambda>0 trust region (ablation) | Grid-TR · ScribbleSup, full scribbles | mIoU | Hurt |
| use a very small trust region (lambda to infinity) instead of moderate lambda (ablation) | Grid-TR · ScribbleSup, full scribbles | mIoU | Hurt |
| use robust KL metric with outlier probability instead of plain cross-entropy (ablation) | Grid-TR · ScribbleSup | mIoU | Improved |
| tune outlier probability epsilon to near the true noise level | robust loss · Fashion-MNIST with noisy labels (half corrupted) | accuracy: best at epsilon=0.4 | Improved |
Closest papers by meaning
- On Regularized Losses for Weakly-supervised CNN Segmentation (ECCV 2018 · similarity 0.91)
- Regularized Loss for Weakly Supervised Single Class Semantic Segmentation (ECCV 2020 · similarity 0.90)
- Regularized Loss With Hyperparameter Estimation for Weakly Supervised Single Class Segmentation (TPAMI 2024 · similarity 0.89)
- Normalized Cut Loss for Weakly-Supervised CNN Segmentation (CVPR 2018 · similarity 0.88)
- DuPL: Dual Student with Trustworthy Progressive Learning for Robust Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.87)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.86)
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.
