Research map / Detection and segmentation / Weakly supervised detection and segmentation
Associating Inter-Image Salient Instances for Weakly Supervised Semantic Segmentation
Ruochen Fan, Qibin Hou, Ming-Ming Cheng, Gang Yu, Ralph R. Martin, Shi-Min Hu
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 graph partitioning on top of the attention module (ablation) | attention-only tagging · PASCAL VOC 2012 test (VOC+SI) | mIoU: 62.7 -> 64.9 | Improved |
| add graph partitioning on top of the attention module (ablation) | attention-only tagging · PASCAL VOC 2012 test (VOC) | mIoU: 62.0 -> 64.0 | Improved |
| add noisy-instance filtering to graph partitioning (ablation) | graph partitioning · PASCAL VOC 2012 test (VOC) | mIoU: 64.0 -> 64.5 | Improved |
| add noisy-instance filtering to graph partitioning (ablation) | graph partitioning · PASCAL VOC 2012 test (simple ImageNet) | mIoU: 62.8 -> 63.9 | Improved |
| add noisy-instance filtering to graph partitioning (ablation) | graph partitioning · PASCAL VOC 2012 test (VOC+SI) | mIoU: 64.9 -> 65.6 | Improved |
| replace random keyword assignment by attention-module tagging (ablation) | salient instance tagging · PASCAL VOC 2012 test (VOC) | mIoU: 56.4 -> 62.0 | Improved |
| add attention probability term (beta=30) to graph partitioning (ablation) | graph partitioning · PASCAL VOC 2012 val | mIoU: 63.2 -> 64.5 | Improved |
| increase beta too far (300 vs 30) (ablation) | graph partitioning · PASCAL VOC 2012 val | mIoU: 64.5 -> 62.9 | Hurt |
| filter low-confidence instances with retention ratio r=0.95 (ablation) | graph partitioning · PASCAL VOC 2012 val | mIoU: 63.8 -> 64.5 | Improved |
| lower retention ratio to r=0.75 (ablation) | graph partitioning · PASCAL VOC 2012 val | mIoU: 64.5 -> 60.9 | Hurt |
| add 24,000 simple ImageNet images to VOC training | ours ResNet101 · PASCAL VOC 2012 test | mIoU: 64.5 -> 65.6 | Improved |
| use ResNet101 instead of VGG16 for the segmentation network | ours (VOC) · PASCAL VOC 2012 test | mIoU: 62.1 -> 64.5 | Improved |
Closest papers by meaning
- Leveraging Instance-, Image- and Dataset-Level Information for Weakly Supervised Instance Segmentation (TPAMI 2020 · similarity 0.92)
- Exploiting Saliency for Object Segmentation From Image Level Labels (CVPR 2017 · similarity 0.90)
- Non-Salient Region Object Mining for Weakly Supervised Semantic Segmentation (CVPR 2021 · similarity 0.90)
- STC: A Simple to Complex Framework for Weakly-Supervised Semantic Segmentation (TPAMI 2016 · similarity 0.90)
- Learning to Detect Instance-Level Salient Objects Using Complementary Image Labels (IJCV 2022 · similarity 0.90)
- Combining Bottom-Up, Top-Down, and Smoothness Cues for Weakly Supervised Image Segmentation (CVPR 2017 · 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.
