Research map / Detection and segmentation / CNN features
CoLA: Conditional Dropout and Language-driven Robust Dual-modal Salient Object Detection
Shuang Hao, Chunlin Zhong, He Tang
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 language-driven quality assessment (LQA) to the baseline (ablation) | CoLA baseline · VT5000 modality complete | F_beta: .801 -> .834 | Improved |
| add Conditional Dropout (CD) to the baseline (ablation) | CoLA baseline · VT5000 average over three conditions | F_beta: .774 -> .793 | Improved |
| combine LQA and CD (ablation) | CoLA baseline · VT5000 average | F_beta: .792 -> .813 (over LQA alone) | Improved |
| apply plain modality dropout instead of CD (ablation) | CoLA with LQA · VT5000 modality complete | F_beta: .834 -> .805 | Hurt |
| freeze the original encoder while training the copied encoder (ablation) | Conditional Dropout · VT5000 modality complete | F_beta: .822 -> .843 | Improved |
| copy and freeze encoder without modality dropout (ablation) | Conditional Dropout · VT5000 missing RGB | F_beta: .754 -> .743 | No clear effect |
| extra 60 epochs of training or only copying the encoder (ablation) | CoLA · VT821 modality complete | F_beta: .839 -> .806 (extra epochs) | Hurt |
| use LQA instead of fixed quality assessors (BRISQUE, GIE, CLIP-IQA) (ablation) | dual-modal fusion · VT5000 | F_beta: .819 (GIE) -> .834 (LQA) | Improved |
| plug Conditional Dropout into LSNet and T-Net | LSNet, T-Net · VT5000 | robustness under missing modality | Improved |
| use larger backbone (ResNet-50 to Swin-B) (ablation) | CoLA · VT5000 modality complete | F_beta: 0.843 -> 0.857 | Improved |
Closest papers by meaning
- Taming Cascaded Mixture-of-Experts for Modality-missing Multi-modal Salient Object Detection (AAAI 2026 · similarity 0.90)
- RA-SOD: Reliability-Aware RGB-T Salient Object Detection under Modality Degradation (ECCV 2026 · similarity 0.88)
- DiMSOD: A Diffusion-Based Framework for Multi-Modal Salient Object Detection (AAAI 2025 · similarity 0.88)
- Specificity-Preserving RGB-D Saliency Detection (ICCV 2021 · similarity 0.88)
- Siamese Network for RGB-D Salient Object Detection and Beyond (TPAMI 2021 · similarity 0.87)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · similarity 0.85)
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.
