Research map / Detection and segmentation / CNN features
Multi-Type Self-Attention Guided Degraded Saliency Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 recurrent local self-attention (RLS) (ablation) | encoder-decoder VGG baseline · HKU-IS (clear) | F-beta: .826 -> .869 | Helped |
| add global pixel self-attention (GPA) (ablation) | encoder-decoder VGG baseline · HKU-IS (clear) | F-beta: .826 -> .845 | Helped |
| add GPA on top of RLS (ablation) | En-De-V+RLS · HKU-IS (clear) | F-beta: .869 -> .896 | Helped |
| add attention transfer network (ATN) instead of adversarial training (ablation) | En-De-V+RLS+GPA · HKU-IS (LR r=4) | F-beta: .778 -> .795 | Helped |
| use MSA structure instead of Amulet under same adversarial training (ablation) | adv-Amulet · HKU-IS (LR r=4) | F-beta: .764 -> .778 | Helped |
| adversarial fine-tuning with degraded images | Amulet · DUT-OMRON/DUTS-TE degraded | F-beta: 0.5490 -> 0.6928 (haze 0.10) | Helped |
| increase ACG recurrent steps beyond 3 (ablation) | ACG in RLS · ECSSD | MAE: 0.0497 -> 0.0578 (t=3 to t=6) | Hurt |
| MSANet vs best baseline on clear images | MSANet · DUTS-TE original | F-beta: 0.8132 (CPDNet) vs 0.7762 | Hurt |
| MSANet on degraded images | CPDNet/AFNet · DUT-OMRON low resolution | F-beta: 0.4241 -> 0.6314 (vs CPDNet) | Helped |
Closest papers by meaning
- Progressive Attention Guided Recurrent Network for Salient Object Detection (CVPR 2018 · similarity 0.88)
- PiCANet: Learning Pixel-Wise Contextual Attention for Saliency Detection (CVPR 2018 · similarity 0.88)
- Non-Local Deep Features for Salient Object Detection (CVPR 2017 · similarity 0.87)
- Deep Saliency With Encoded Low Level Distance Map and High Level Features (CVPR 2016 · similarity 0.87)
- ICNet: Intra-saliency Correlation Network for Co-Saliency Detection (NeurIPS 2020 · similarity 0.87)
- Deep Unsupervised Saliency Detection: A Multiple Noisy Labeling Perspective (CVPR 2018 · similarity 0.87)
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.