CXR Vision-Language Models
Collection
Fine-tunes of Qwen3.5-9B for chest X-ray report generation, plus RF-DETR detection and UNet segmentation baselines. • 10 items • Updated
ChestX-Det Segmentation — UNet + EfficientNet-B4
13-class CXR segmentation baseline. 40 epochs, mean_dice=0.3091.
| Class | Dice | Class | Dice |
|---|---|---|---|
| Atelectasis | 0.34 | Fibrosis | 0.14 |
| Calcification | 0.12 | Fracture | 0.37 |
| Cardiomegaly | 0.05 | Mass | 1.00* |
| Consolidation | 0.03 | Nodule | 0.06 |
| Diffuse Nodule | 0.05 | Pleural Thickening | 0.00 |
| Effusion | 0.02 | Pneumothorax | 1.00* |
| Emphysema | 0.06 | ||
| *Likely degenerate on small val set. |
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support