Research map / Detection and segmentation / Weakly supervised detection and segmentation
Seed, Expand and Constrain: Three Principles for Weakly-Supervised Image Segmentation
Alexander Kolesnikov, Christoph H. Lampert
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 |
|---|---|---|---|
| use GWRP instead of global max pooling for present classes (ablation) | SEC · PASCAL VOC 2012 val | mIoU (%): 47.3 -> 50.7 | Improved |
| use GWRP instead of global average pooling for present classes (ablation) | SEC · PASCAL VOC 2012 val | mIoU (%): 45.1 -> 50.7 | Improved |
| add expansion and constrain losses to seeding loss (full SEC vs L_seed only) (ablation) | SEC · PASCAL VOC 2012 val | mIoU (%): 49.2 -> 50.7 | Improved |
| add constrain-to-boundary loss to seeding loss (ablation) | SEC · PASCAL VOC 2012 val | mIoU (%): 49.4 -> 50.7 | Improved |
| add expansion loss to seeding loss alone (ablation) | SEC · PASCAL VOC 2012 val | mIoU (%): 49.2 -> 45.7 | Hurt |
| add constrain loss to seeding loss alone (ablation) | SEC · PASCAL VOC 2012 val | mIoU (%): 49.2 -> 49.4 | No clear effect |
| add seeding loss to expansion and constrain losses (ablation) | SEC · PASCAL VOC 2012 val | mIoU (%): 17.2 -> 50.7 | Improved |
| add seeding loss to expansion loss alone (ablation) | SEC · PASCAL VOC 2012 val | mIoU (%): 27.8 -> 45.7 | Improved |
| use large field-of-view network (378x378) instead of small (211x211) (ablation) | SEC · PASCAL VOC 2012 val | mIoU (%) | Improved |
| train SEC with image-level labels versus prior best weakly supervised (val) | SEC · PASCAL VOC 2012 val | mIoU (%): 36.6 -> 50.7 | Improved |
| train SEC with image-level labels versus prior best weakly supervised (test) | SEC · PASCAL VOC 2012 test | mIoU (%): 38.0 -> 51.7 | Improved |
Closest papers by meaning
- Regularized Loss for Weakly Supervised Single Class Semantic Segmentation (ECCV 2020 · similarity 0.93)
- Combining Bottom-Up, Top-Down, and Smoothness Cues for Weakly Supervised Image Segmentation (CVPR 2017 · similarity 0.92)
- Revisiting Dilated Convolution: A Simple Approach for Weakly- and Semi-Supervised Semantic Segmentation (CVPR 2018 · similarity 0.92)
- Normalized Cut Loss for Weakly-Supervised CNN Segmentation (CVPR 2018 · similarity 0.91)
- Weakly-Supervised Semantic Segmentation Network With Deep Seeded Region Growing (CVPR 2018 · similarity 0.91)
- Webly Supervised Semantic Segmentation (CVPR 2017 · similarity 0.91)
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.
