Research map / Detection and segmentation / Weakly supervised detection and segmentation
Incorporating Network Built-in Priors in Weakly-Supervised Semantic Segmentation
Fatemeh Sadat Saleh, Mohammad Sadegh Aliakbarian, Mathieu Salzmann, Lars Petersson, José Manuel Álvarez, Stephen Jay Gould
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 foreground/background masks from pretrained VGG-16 activations (ablation) | tag-only segmentation network · PASCAL VOC 2012 val | mIoU: 31.0 -> 47.3 | Improved |
| add multi-class masks (localization + fusion) instead of no mask (ablation) | tag-only segmentation network · PASCAL VOC 2012 val | mIoU: 31.0 -> 49.2 | Improved |
| add higher-order CRF term to fg/bg masks (ablation) | fg/bg mask method · PASCAL VOC 2012 val | mIoU: 47.3 -> 48.0 | Improved |
| add higher-order CRF term to localization masks (ablation) | localization prior method · PASCAL VOC 2012 val | mIoU: 45.9 -> 46.6 | Improved |
| add fusion with fg/bg prior to localization masks (ablation) | localization prior method · PASCAL VOC 2012 val | mIoU: 45.9 -> 49.2 | Improved |
| use large field of view instead of small (ablation) | multi-class mask method with higher-order term · PASCAL VOC 2012 val | mIoU: 49.3 -> 50.9 | Improved |
| use built-in fg/bg masks instead of external objectness masks | fg/bg mask generation · PASCAL VOC 2012 val | mean IoU of masks: 52.34% -> 60.08% | Improved |
| add fusion to localization masks used directly without training (ablation) | multi-class masks · PASCAL VOC 2012 val | mIoU: 43.0 -> 46.6 | Improved |
| train a network on masks instead of predicting from masks directly | multi-class masks · PASCAL VOC 2012 val | mIoU: 46.6 -> 50.9 | Improved |
| use multi-class masks instead of fg/bg masks | Ours · PASCAL VOC 2012 test | mIoU: 49.6 -> 52.6 | Improved |
| use our method instead of SEC | weakly supervised segmentation · MS COCO val | mIoU: 22.4 -> 20.4 | Hurt |
| use our method instead of SEC | weakly supervised segmentation · YTO | mIoU: 48.6 -> 51.6 | Improved |
Closest papers by meaning
- Built-in Foreground/Background Prior for Weakly-Supervised Semantic Segmentation (ECCV 2016 · similarity 0.98)
- Weakly-Supervised Semantic Segmentation by Iteratively Mining Common Object Features (CVPR 2018 · similarity 0.92)
- Combining Bottom-Up, Top-Down, and Smoothness Cues for Weakly Supervised Image Segmentation (CVPR 2017 · similarity 0.91)
- Weakly Supervised Semantic Segmentation via Box-Driven Masking and Filling Rate Shifting (TPAMI 2023 · similarity 0.90)
- Railroad Is Not a Train: Saliency As Pseudo-Pixel Supervision for Weakly Supervised Semantic Segmentation (CVPR 2021 · similarity 0.90)
- Progressive Feature Self-Reinforcement for Weakly Supervised Semantic Segmentation (AAAI 2024 · similarity 0.89)
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.
