Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- chest-xray
|
| 4 |
+
- radiology
|
| 5 |
+
- visual-question-answering
|
| 6 |
+
- differential-vqa
|
| 7 |
+
- mimic-cxr
|
| 8 |
+
license: apache-2.0
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# LAPVQA — Differential VQA (Frozen Off-the-shelf Encoders)
|
| 12 |
+
|
| 13 |
+
Part of the [LAPVQA collection](https://huggingface.co/collections/dmusingu/lapvqa).
|
| 14 |
+
|
| 15 |
+
## Description
|
| 16 |
+
|
| 17 |
+
Task heads for **Differential VQA (DiffVQA)**: given a *prior* and a *current* chest X-ray,
|
| 18 |
+
answer natural-language questions about radiological changes between the two studies.
|
| 19 |
+
Trained on MIMIC-Diff-VQA with five **frozen** off-the-shelf vision encoders.
|
| 20 |
+
|
| 21 |
+
## Results (test set)
|
| 22 |
+
|
| 23 |
+
| Encoder | BLEU-1 | BLEU-4 | ROUGE-1 | RadGraph-s |
|
| 24 |
+
|---|---|---|---|---|
|
| 25 |
+
| CLIP ViT-L/14 | 0.184 | 0.128 | 0.336 | 0.322 |
|
| 26 |
+
| CoCa | 0.196 | 0.138 | 0.320 | 0.317 |
|
| 27 |
+
| Florence-2 | 0.191 | 0.138 | 0.319 | 0.318 |
|
| 28 |
+
| SigLIP | 0.186 | 0.131 | 0.322 | 0.313 |
|
| 29 |
+
| OWLv2 | — | — | — | — |
|
| 30 |
+
|
| 31 |
+
## Files
|
| 32 |
+
|
| 33 |
+
| File | Encoder backbone |
|
| 34 |
+
|---|---|
|
| 35 |
+
| `clip-vit-l14_best.pt` | CLIP ViT-L/14 |
|
| 36 |
+
| `coca_best.pt` | CoCa |
|
| 37 |
+
| `florence2_best.pt` | Florence-2 |
|
| 38 |
+
| `siglip_best.pt` | SigLIP |
|
| 39 |
+
| `owlv2_best.pt` | OWLv2 |
|