Research map / Detection and segmentation / Weakly supervised detection and segmentation
Built-in Foreground/Background Prior for Weakly-Supervised Semantic Segmentation
Fatemehsadat Saleh, Mohammad Sadegh Aliakbarian, Mathieu Salzmann, Lars Petersson, Stephen Gould, Jose M. Alvarez
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 the built-in foreground/background mask to the loss (ablation) | weakly-supervised segmentation (tags) · PASCAL VOC 2012 val (image tags) | mIoU (%): 31.0 -> 44.8 | Improved |
| add the built-in mask on MIRFLICKR-1M training (ablation) | weakly-supervised segmentation (tags) · PASCAL VOC 2012 val (trained on MIRFLICKR-1M subset) | mIoU (%): 30.6 -> 38.1 | Improved |
| use built-in mask instead of external objectness prior | weakly-supervised segmentation · PASCAL VOC 2012 val (image tags) | mIoU (%): 32.2 -> 44.8 | Improved |
| add dense CRF smoothing | Ours (Tags) · PASCAL VOC 2012 val (image tags) | mIoU (%): 44.8 -> 46.6 | Improved |
| add dense CRF smoothing | Ours (CheckMask) · PASCAL VOC 2012 val (additional weak supervision) | mIoU (%): 49.0 -> 51.5 | Improved |
| switch to the final loss formulation instead of the Eq. variant (ablation) | Ours (Tags) · PASCAL VOC 2012 val (image tags) | mIoU (%): 44.8 -> 44.8 | No clear effect |
| add CheckMask supervision (choose best mask among candidates) | Ours (Tags) · PASCAL VOC 2012 val (image tags) with CRF | mIoU (%): 4.9 mIOU point | Improved |
| add CheckMask supervision | Ours (Tags) · PASCAL VOC 2012 val (image tags) | mIoU (%): 4.2 mIOU point | Improved |
| add CheckMask supervision | Ours (Tags) · PASCAL VOC 2012 val (trained on MIRFLICKR-1M subset) | mIoU (%): 38.1 -> 43.9 | Improved |
| use Ours instead of CCNN | MIRFLICKR-1M-trained model · PASCAL VOC 2012 val (trained on MIRFLICKR-1M subset) | mIoU (%): 30.5 -> 38.1 | Improved |
| use Ours instead of CCNN on rare class chair | MIRFLICKR-1M-trained model · Chair class IoU, PASCAL VOC 2012 val (trained on MIRFLICKR-1M subset) | mIoU (%): 10.7 -> 17.0 | Improved |
| use Ours instead of CCNN on rare class sofa | MIRFLICKR-1M-trained model · Sofa class IoU, PASCAL VOC 2012 val (trained on MIRFLICKR-1M subset) | mIoU (%): 0 -> 16.8 | Improved |
Closest papers by meaning
- Incorporating Network Built-in Priors in Weakly-Supervised Semantic Segmentation (TPAMI 2017 · similarity 0.98)
- Combining Bottom-Up, Top-Down, and Smoothness Cues for Weakly Supervised Image Segmentation (CVPR 2017 · similarity 0.92)
- Weakly-Supervised Semantic Segmentation by Iteratively Mining Common Object Features (CVPR 2018 · similarity 0.92)
- Railroad Is Not a Train: Saliency As Pseudo-Pixel Supervision for Weakly Supervised Semantic Segmentation (CVPR 2021 · similarity 0.91)
- Reliability Does Matter: An End-to-End Weakly Supervised Semantic Segmentation Approach (AAAI 2020 · 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.
