BLIP — Chest X-Ray Captioning (MIMIC-CXR Fine-Tune)
A fine-tuned version of Salesforce/blip-image-captioning-base
on the MIMIC-CXR dataset for automatic chest X-ray report generation.
Dataset
- Source:
itsanmolgupta/mimic-cxr-dataset - Train split: 19,600 samples (indices 9,000 – 28,600)
- Validation split: 2,000 samples (indices 28,600 – 30,600)
Usage
from transformers import AutoProcessor, AutoModelForImageTextToText
from PIL import Image
import torch
repo_id = "AliFadel/blip-cxr-mimic-finetuned"
processor = AutoProcessor.from_pretrained(repo_id)
model = AutoModelForImageTextToText.from_pretrained(repo_id)
model.eval()
image = Image.open("chest_xray.jpg").convert("RGB")
inputs = processor(images=image, text="Chest X-Ray", return_tensors="pt")
with torch.no_grad():
ids = model.generate(
**inputs,
num_beams=5,
max_length=150,
repetition_penalty=1.5,
top_p=0.95,
)
caption = processor.batch_decode(ids, skip_special_tokens=True)[0]
print(caption)
Intended Use
This model is intended for research purposes only. It should not be used as a clinical diagnostic tool without expert medical supervision.