Research map / Label-efficient and robust learning / Recognition and long-tail
GLoRIA: A Multimodal Global-Local Representation Learning Framework for Label-Efficient Medical Image Recognition
Shih-Cheng Huang, Liyue Shen, Matthew P. Lungren, Serena Yeung
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 local representations to global at inference (ablation) | GLoRIA · CheXpert 5x200 retrieval | Prec@5: 67.02 -> 69.24 | Improved |
| add local representations to global at inference (ablation) | GLoRIA · CheXpert 5x200 retrieval | Prec@100: 49.55 -> 53.78 | Improved |
| add global representations to local only (ablation) | GLoRIA · CheXpert 5x200 retrieval | Prec@10: 64.58 -> 67.22 | Improved |
| add local contrastive loss to global-only contrastive training (vs ConVIRT) with global features | GLoRIA · CheXpert 5x200 retrieval | Prec@5: 66.98 -> 67.02 | No clear effect |
| use GLoRIA instead of ConVIRT | GLoRIA · RSNA fine-tuned classification (1% labels) | AUROC (%): 77.4 -> 86.1 | Improved |
| use GLoRIA instead of ConVIRT | GLoRIA · CheXpert fine-tuned classification (1% labels) | AUROC (%): 85.9 -> 86.6 | Improved |
| use GLoRIA 1% labels instead of ImageNet init with 100% labels | GLoRIA · CheXpert fine-tuned classification | AUROC (%): 81.4 -> 86.6 | Improved |
| use GLoRIA initialization instead of ConVIRT | GLoRIA · SIIM pneumothorax segmentation (1% labels) | Dice: 0.250 -> 0.358 | Improved |
| use GLoRIA initialization instead of ConVIRT | GLoRIA · SIIM pneumothorax segmentation (10% labels) | Dice: 0.432 -> 0.469 | Improved |
| use GLoRIA instead of ImageNet initialization with full labels | GLoRIA · SIIM pneumothorax segmentation (100% labels) | Dice: 0.635 -> 0.634 | No clear effect |
| use zero-shot GLoRIA instead of fine-tuned classifier with 100% labels | GLoRIA · CheXpert 5x200 classification | F1: 0.63 -> 0.67 | Improved |
| use zero-shot GLoRIA instead of fine-tuning on 1% labels | GLoRIA · RSNA classification | F1: 0.57 -> 0.58 | No clear effect |
Later papers that compared with this method
2 later accepted papers report GLoRIA as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Integrating Prior Knowledge in Contrastive Learning with Kernel (ICML 2023 · better on 1 of 2 shared results)
- Advancing Radiograph Representation Learning with Masked Record Modeling (ICLR 2023 · 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
- Unlocking the Power of Spatial and Temporal Information in Medical Multimodal Pre-training (ICML 2024 · similarity 0.86)
- G2D: From Global to Dense Radiography Representation Learning via Vision-Language Pre-training (NeurIPS 2024 · similarity 0.86)
- Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding (ICLR 2025 · similarity 0.86)
- Autoregressive Sequence Modeling for 3D Medical Image Representation (AAAI 2025 · similarity 0.85)
- MedMO: Grounding and Understanding Multimodal Large Language Model for Medical Images (CVPR 2026 · similarity 0.85)
- Interpretable Bilingual Multimodal Large Language Model for Diverse Biomedical Tasks (ICLR 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.
