Research map / Detection and segmentation / Weakly supervised detection and segmentation
RefCLIP: A Universal Teacher for Weakly Supervised Referring Expression Comprehension
Lei Jin, Gen Luo, Yiyi Zhou, Xiaoshuai Sun, Guannan Jiang, Annan Shu, Rongrong Ji
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 |
|---|---|---|---|
| Use anchor-text contrastive learning with filtered anchors instead of detected-region ranking | RefCLIP · RefCOCO val | IoU@0.5 accuracy: 38.08 -> 60.36 | Improved |
| Use RefCLIP to pseudo-label MAttNet training | MAttNet · RefCOCO testB | IoU@0.5 accuracy: 38.59 -> 71.27 | Improved |
| Remove scale selection and confidence filtering (ablation) | RefCLIP · RefCOCO val | IoU@0.5 accuracy: 33.71 -> 60.36 | Hurt |
| Select anchors from one scale (ablation) | RefCLIP · RefCOCO val | IoU@0.5 accuracy: 33.71 -> 50.75 | Improved |
| Add confidence-based anchor filtering (ablation) | RefCLIP · RefCOCO val | IoU@0.5 accuracy: 50.75 -> 53.30 | Improved |
| Add negative anchor augmentation (ablation) | RefCLIP · RefCOCO val | IoU@0.5 accuracy: 53.30 -> 60.36 | Improved |
| Use the 52 × 52 anchor scale (ablation) | RefCLIP · RefCOCO val | IoU@0.5 accuracy: 13 × 13: 60.36; 52 × 52: 0.00 | Hurt |
| Use the 13 × 13 anchor scale rather than 26 × 26 (ablation) | RefCLIP · RefCOCO val | IoU@0.5 accuracy: 11.23 -> 60.36 | Improved |
| Keep the top 20% of anchors by confidence (ablation) | RefCLIP · RefCOCO val | IoU@0.5 accuracy: 20.84 -> 59.31 | Improved |
| Keep only the top 5% of anchors by confidence (ablation) | RefCLIP · RefCOCO val | IoU@0.5 accuracy: 60.36 -> 48.46 | Hurt |
| Add data augmentation to pseudo-label student training (ablation) | RealGIN · RefCOCO val | IoU@0.5 accuracy: 57.36 -> 58.99 | Improved |
| Add EMA to student training with data augmentation (ablation) | RealGIN · RefCOCO val | IoU@0.5 accuracy: 58.99 -> 59.43 | Improved |
Later papers that compared with this method
1 later accepted paper reports RefCLIP as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- APL: Anchor-based Prompt Learning for One-stage Weakly Supervised Referring Expression Comprehension (ECCV 2024 · better on 6 of 7 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
- Rethinking Two-Stage Referring Expression Comprehension: A Novel Grounding and Segmentation Method Modulated by Point (AAAI 2024 · similarity 0.88)
- Revisiting Counterfactual Problems in Referring Expression Comprehension (CVPR 2024 · similarity 0.88)
- Ref-Adv: Exploring MLLM Visual Reasoning in Referring Expression Tasks (ICLR 2026 · similarity 0.88)
- RefDetector: A Simple Yet Effective Matching-based Method for Referring Expression Comprehension (AAAI 2025 · similarity 0.87)
- DViN: Dynamic Visual Routing Network for Weakly Supervised Referring Expression Comprehension (CVPR 2025 · similarity 0.87)
- DiffusionREC: Diffusion Model with Adaptive Condition for Referring Expression Comprehension (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.
