Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Towards Zero-Shot Diabetic Retinopathy Grading: Learning Generalized Knowledge via Prompt-Driven Matching and Emulating
Huan Wang, Haoran Li, Yuxin Lin, Huaming Chen, Jun Yan, Lijuan Wang, Jiahua Shi, Qihao Xu, Yongting Hu, Yong Xu, Jun Shen
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 |
|---|---|---|---|
| two-stage prompt matching then emulating (ProME-DR) instead of fully fine-tuned SigLIP | SigLIP-B/16 · Seen-to-Unseen DR grades, 8-dataset average (16-shot) | HM accuracy: 67.69 -> 71.81 | Improved |
| ProME-DR vs prompt-tuning baselines on unseen classes | MMRL · Seen-to-Unseen DR grades, 8-dataset average (16-shot) | unseen-class accuracy: 39.37 -> 68.04 | Improved |
| prompt tuning (VPT) instead of fully fine-tuned CLIP | CLIP-B/32 · Seen-to-Unseen DR grades, 8-dataset average (16-shot) | HM accuracy: 65.64 -> 50.72 | Hurt |
| most prompt-based baselines versus fully fine-tuned CLIP on unseen classes | prompt tuning (e.g. MMRL) · Seen-to-Unseen DR grades, 8-dataset average (16-shot) | unseen-class accuracy: 47.31 -> 39.37 | Hurt |
| train on MFIDDR and test directly on 7 other datasets using ProME-DR | SigLIP-B/16 · Cross-dataset generalization | accuracy: 55.97 -> 64.16 | Improved |
| ProME-DR vs SMVDR supervised on MFIDDR | SMVDR · MFIDDR supervised | accuracy: 81.64 -> 87.90 | Improved |
| ProME-DR vs SMVDR on specificity | SMVDR · MFIDDR supervised | specificity: 89.92 -> 85.29 | Hurt |
| add contrastive loss Lcl (lambda2 > 0) and distillation loss Ldis (lambda1 > 0) (ablation) | ProME-DR · Seen-to-Unseen DR grades | HM accuracy | Improved |
| increase temperature tau toward 0.06 (ablation) | ProME-DR · Seen-to-Unseen DR grades | HM accuracy | Improved |
Closest papers by meaning
- Like an Ophthalmologist: Dynamic Selection Driven Multi-View Learning for Diabetic Retinopathy Grading (AAAI 2025 · similarity 0.91)
- FOCUS: Benchmarking Retinal Model Generalization from Foundation Vision Encoders to Multimodal LLMs (NeurIPS 2026 · similarity 0.89)
- Domain Adaptive Diabetic Retinopathy Grading with Model Absence and Flowing Data (CVPR 2025 · similarity 0.88)
- HACDR-Net: Heterogeneous-Aware Convolutional Network for Diabetic Retinopathy Multi-Lesion Segmentation (AAAI 2024 · similarity 0.86)
- Prompt as Knowledge Bank: Boost Vision-language model via Structural Representation for zero-shot medical detection (ICLR 2025 · similarity 0.84)
- Deep Hierarchies and Invariant Disease-Indicative Feature Learning for Computer Aided Diagnosis of Multiple Fundus Diseases (AAAI 2025 · similarity 0.84)
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.
