Research map / Detection and segmentation / Weakly supervised detection and segmentation
Weakly-Supervised Semantic Segmentation by Iterative Affinity Learning
Xiang Wang, Sifei Liu, Huimin Ma, Ming-Hsuan Yang
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 mining of confident regions to supervise the pairwise network (ablation) | Iterative affinity learning · PASCAL VOC 2012 train (VGG16), step 5 | mIoU: 56.4 -> 63.1 | Improved |
| add mining of confident regions (ablation) | Iterative affinity learning · PASCAL VOC 2012 train (VGG16), 500-image energy, step 5 | energy: 0.048 -> 0.029 | Improved |
| add the pairwise affinity network (ablation) | Iterative affinity learning · PASCAL VOC 2012 train (VGG16), step 5 | mIoU: 54.3 -> 63.1 | Improved |
| add the pairwise affinity network (ablation) | Iterative affinity learning · PASCAL VOC 2012 val (VGG16) | mIoU: lower than our method by 9.2% | Improved |
| add region smoothness loss (ablation) | Pairwise affinity network · PASCAL VOC 2012 val (VGG16) | mIoU: 61.2 -> 62.0 | Improved |
| iterate affinity learning for 5 steps | Pairwise network · PASCAL VOC 2012 train (VGG16) | mIoU: 55.2 -> 63.1 | Improved |
| iterate affinity learning for 5 steps | Unary network · PASCAL VOC 2012 train (VGG16) | mIoU: 51.5 -> 61.7 | Improved |
| iterate beyond step 5 | Pairwise network · PASCAL VOC 2012 train (VGG16) | mIoU: 63.1 -> 63.2 | No clear effect |
| refine affinity network with confident regions instead of PSA-style regions, iteratively | Iterative PSA · PASCAL VOC 2012 train (VGG16), step 5 | mIoU: 48.1 -> 63.1 | Improved |
| use our confidence-network mining instead of PSA mining for confident-region precision | Iterative PSA · PASCAL VOC 2012 train (VGG16), step 5 | precision: 68.7 -> 81.2 | Improved |
| apply the learned pairwise refinement to unary output | Unary network · PASCAL VOC 2012 train (VGG16), step 1 | mIoU: 51.5 -> 55.2 | Improved |
| replace CAM seeds with first-step unary output | CAM seeds · PASCAL VOC 2012 train (VGG16) | mIoU: 46.2 -> 51.5 | Improved |
Closest papers by meaning
- Learning Pixel-Level Semantic Affinity With Image-Level Supervision for Weakly Supervised Semantic Segmentation (CVPR 2018 · similarity 0.92)
- Learning Affinity From Attention: End-to-End Weakly-Supervised Semantic Segmentation With Transformers (CVPR 2022 · similarity 0.92)
- Label-efficient Segmentation via Affinity Propagation (NeurIPS 2023 · similarity 0.91)
- Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning (ICLR 2021 · similarity 0.91)
- CIAN: Cross-Image Affinity Net for Weakly Supervised Semantic Segmentation (AAAI 2020 · similarity 0.90)
- Weakly-Supervised Semantic Segmentation by Iteratively Mining Common Object Features (CVPR 2018 · 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.
