Create app.py

app.py

@@ -0,0 +1,41 @@
+# app.py
+import gradio as gr
+import json
+from transformers import pipeline
+
+# Load labels from external JSON
+with open("labels.json", "r", encoding="utf-8") as f:
+    candidate_labels = json.load(f)
+
+# Initialize the MedSigLIP pipeline
+pipe = pipeline("zero-shot-image-classification", model="google/medsiglip-448")
+
+# Define prediction function
+def classify_image(image):
+    result = pipe(image, candidate_labels=candidate_labels)
+    # Convert result list to {label: score}
+    formatted = {item["label"]: round(item["score"] * 100, 2) for item in result}
+    # Sort descending by confidence
+    sorted_result = dict(sorted(formatted.items(), key=lambda x: x[1], reverse=True))
+    return sorted_result
+
+# Build Gradio interface
+demo = gr.Interface(
+    fn=classify_image,
+    inputs=gr.Image(type="filepath", label="Upload Chest X-ray or Medical Image"),
+    outputs=gr.Label(num_top_classes=5, label="Top Predictions"),
+    title="🩻 MedSigLIP Zero-Shot Medical Image Classifier",
+    description=(
+        "This demo uses Google's MedSigLIP (Sigmoid Loss for Language Image Pre-training) model "
+        "for zero-shot medical image classification. Upload an image and the model will estimate "
+        "its similarity with known medical conditions (loaded dynamically from labels.json)."
+    ),
+    allow_flagging="never",
+    examples=[
+        ["https://storage.googleapis.com/dx-scin-public-data/dataset/images/3445096909671059178.png"],
+        ["https://storage.googleapis.com/dx-scin-public-data/dataset/images/-5669089898008966381.png"]
+    ],
+)
+
+if __name__ == "__main__":
+    demo.launch()