Research map / Detection and segmentation / CNN features
MobileSal: Extremely Efficient RGB-D Salient Object Detection
Yu-Huan Wu, Yun Liu, Jun Xu, Jia-Wang Bian, Yu-Chao Gu, Ming‐Ming Cheng
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 implicit depth restoration (IDR) branch, fusing at coarsest level (ablation) | MobileSal · NJU2K test | F-measure max: 0.896 -> 0.906 | Improved |
| fuse RGB and depth at the coarsest level only instead of all levels (ablation) | MobileSal with IDR · NJU2K test | F-measure max: 0.899 -> 0.906 | Improved |
| fuse only at the coarsest level instead of all levels (speed) (ablation) | MobileSal · speed | FPS: 260 -> 450 FPS | Improved |
| use two-stream fusion instead of early fusion single stream (ablation) | MobileSal with IDR · NJU2K test | F-measure max: 0.900 -> 0.906 | Improved |
| add CPR module to RGB backbone (ablation) | MobileSal · NJU2K test | F-measure max: 0.852 -> 0.887 | Improved |
| add depth stream (multiplication fusion) (ablation) | RGB+CPR · NJU2K test | F-measure max: 0.887 -> 0.894 | Improved |
| add CMF module (ablation) | RGB+CPR+Depth · NJU2K test | F-measure max: 0.894 -> 0.906 | Improved |
| element-wise multiplication instead of addition or concatenation for initial fusion (ablation) | CMF · NJU2K test | F-measure max: 0.897 -> 0.906 | Improved |
| keep RGB-D feature T in CMF fusion (ablation) | CMF · NJU2K test | F-measure max: 0.895 -> 0.906 | Improved |
| keep RGB attention vector v in CMF fusion (ablation) | CMF · NJU2K test | F-measure max: 0.902 -> 0.906 | Improved |
| use compact dilation rates 1, 2, 3 instead of sparse 1, 3, 6 (ablation) | CPR · NJU2K test | F-measure max: 0.903 -> 0.906 | Improved |
| use SSIM loss instead of L1/L2 loss for IDR (ablation) | IDR · NJU2K test | F-measure max: 0.5%/0.3% lower with L1/L2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- JLDCF — JL-DCF: Joint Learning and Densely-Cooperative Fusion Framework for RGB-D Salient Object Detection (CVPR 2020) · better on 5 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- RGB-D Salient Object Detection via 3D Convolutional Neural Networks (AAAI 2021 · similarity 0.89)
- Promoting Saliency From Depth: Deep Unsupervised RGB-D Saliency Detection (ICLR 2022 · similarity 0.89)
- A Single Stream Network for Robust and Real-time RGB-D Salient Object Detection (ECCV 2020 · similarity 0.89)
- Accurate RGB-D Salient Object Detection via Collaborative Learning (ECCV 2020 · similarity 0.89)
- DVSOD: RGB-D Video Salient Object Detection (NeurIPS 2023 · similarity 0.88)
- Alignment-Free RGB-T Salient Object Detection: A Large-Scale Dataset and Progressive Correlation Network (AAAI 2025 · similarity 0.86)
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.
