Research map / Detection and segmentation / Weakly supervised detection and segmentation
Referring Image Segmentation Using Text Supervision
Fang Liu, Yuhao Liu, Yuqiu Kong, Ke Xu, Lihe Zhang, Baocai Yin, Gerhard Hancke, Rynson Lau
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 bilateral prompt (both directions) instead of no prompt (ablation) | TRIS Step-1 · RefCOCOg (U) train | PointM: 36.93 -> 40.92 | Improved |
| bilateral prompt instead of unilateral text-to-visual prompt (ablation) | TRIS Step-1 · RefCOCOg (U) train | IoU: 19.70 -> 20.43 | Improved |
| add positive foreground enhancement term in calibration (ablation) | TRIS Step-1 with bilateral prompt · RefCOCOg (U) train | IoU: 20.43 -> 24.18 | Improved |
| add negative background suppression term in calibration (ablation) | TRIS Step-1 with bilateral prompt · RefCOCOg (U) train | IoU: 20.43 -> 22.43 | Improved |
| combine both calibration terms with bilateral prompt (ablation) | TRIS Step-1 · RefCOCOg (U) train | IoU: 18.94 -> 27.81 | Improved |
| add positive response map selection (PRMS) (ablation) | TRIS pseudo-label generation · RefCOCO+ val | IoU: 22.31 -> 24.48 | Improved |
| add PRMS (ablation) | TRIS pseudo-label generation · RefCOCOg val (U) | IoU: 26.62 -> 27.06 | No clear effect |
| use negative text queries (N+1 from 1 to 48) in classification (ablation) | TRIS Step-1 · RefCOCOg (U) train | IoU: 11.16 -> 27.81 | Improved |
| increase negative queries from 48 to 60 (ablation) | TRIS Step-1 · RefCOCOg (U) train | IoU: 27.81 -> 27.61 | Hurt |
| increase calibration negatives K from 6 to 12 (ablation) | TRIS Step-1 · RefCOCOg (U) train | IoU: 27.81 -> 27.63 | Hurt |
| train segmentation network on pseudo-labels (Step-2) instead of Step-1 response maps (ablation) | TRIS · RefCOCOg val (U) | IoU: 26.62 -> 36.19 | Improved |
| text-only supervision versus text-supervised adaptations (WWbL) | WWbL · ReferIt test | IoU: 27.68 -> 44.57 | 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.
- GroupViT — GroupViT: Semantic Segmentation Emerges From Text Supervision (CVPR 2022) · better on 1 of 1 shared result
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
- Boosting Weakly Supervised Referring Image Segmentation via Progressive Comprehension (NeurIPS 2024 · similarity 0.92)
- Curriculum Point Prompting for Weakly-Supervised Referring Image Segmentation (CVPR 2024 · similarity 0.90)
- Mask Grounding for Referring Image Segmentation (CVPR 2024 · similarity 0.90)
- Hybrid Global-Local Representation with Augmented Spatial Guidance for Zero-Shot Referring Image Segmentation (CVPR 2025 · similarity 0.89)
- Prompt-Driven Referring Image Segmentation with Instance Contrasting (CVPR 2024 · similarity 0.89)
- IteRPrimE: Zero-shot Referring Image Segmentation with Iterative Grad-CAM Refinement and Primary Word Emphasis (AAAI 2025 · 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.
