Research map / Detection and segmentation / Weakly supervised detection and segmentation
Promoting Saliency From Depth: Deep Unsupervised RGB-D 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 |
|---|---|---|---|
| Use iterative depth-disentangled pseudo-label updates and attentive training instead of initial pseudo-label training alone | CPD-based unsupervised RGB-D SOD · NLPR | Fβ: .570 -> .745 | Helped |
| Use iterative depth-disentangled pseudo-label updates and attentive training instead of initial pseudo-label training alone | CPD-based unsupervised RGB-D SOD · NJUD | Fβ: .627 -> .719 | Helped |
| Add attentive training strategy to the backbone (ablation) | Backbone saliency network · NJUD and NLPR | Fβ: +3% on NJUD; +5.8% on NLPR | Helped |
| Add the full DSU strategy to the backbone (ablation) | Backbone saliency network · NJUD and NLPR | Fβ: +11.3% on NJUD; +23.5% on NLPR | Helped |
| Add the full DSU strategy to the backbone (ablation) | Backbone saliency network · NJUD and NLPR | MAE: reducing MAE by 22.4% and 41.9% | Helped |
| Remove the consistency loss (ablation) | DSU · NJUD and NLPR | performance | Hurt |
| Remove holistic attention (ablation) | DSU · NJUD and NLPR | performance | Hurt |
| Remove DLU and retain only CRF for pseudo-label refinement (ablation) | DSU · NJUD and NLPR | performance | Hurt |
| Add CRF refinement to initial pseudo-labels (ablation) | Initial pseudo-label generation · NJUD | MAE: 0.162 -> 0.144 | Helped |
| Replace CRF-only refinement with the proposed updated training labels (ablation) | Pseudo-label generation · NJUD | MAE: 0.144 -> 0.116 | Helped |
| Use raw depth map rather than learned disentangled depth cues (ablation) | DSU pseudo-label update · NLPR and NJUD | Fβ: 0.164 Fβ for depth map; 0.429 Fβ for depth map + OTSU (NJUD) | Hurt |
| Add CRF smoothing to saliency-guided depth (ablation) | DSU · NLPR and NJUD | Fβ: 0.745 -> 0.751 (NLPR); 0.719 -> 0.724 (NJUD) | Helped |
Closest papers by meaning
- Deep RGB-D Saliency Detection With Depth-Sensitive Attention and Automatic Multi-Modal Fusion (CVPR 2021 · similarity 0.92)
- DVSOD: RGB-D Video Salient Object Detection (NeurIPS 2023 · similarity 0.92)
- RGB-D Salient Object Detection via 3D Convolutional Neural Networks (AAAI 2021 · similarity 0.92)
- Joint Semantic Mining for Weakly Supervised RGB-D Salient Object Detection (NeurIPS 2021 · similarity 0.92)
- Alignment-Free RGB-T Salient Object Detection: A Large-Scale Dataset and Progressive Correlation Network (AAAI 2025 · similarity 0.89)
- DiMSOD: A Diffusion-Based Framework for Multi-Modal Salient Object Detection (AAAI 2025 · 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.