Create app.py

app.py

@@ -0,0 +1,43 @@
+from fastapi import FastAPI, UploadFile, File
+from fastapi.responses import JSONResponse
+from PIL import Image as PILImage
+from transformers import AutoImageProcessor, SiglipForImageClassification
+import torch
+import io
+import warnings
+
+MODEL_IDENTIFIER = "Ateeqq/ai-vs-human-image-detector"
+DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+# Suppress warnings
+warnings.filterwarnings("ignore", message="Possibly corrupt EXIF data.")
+
+# Load processor and model once
+processor = AutoImageProcessor.from_pretrained(MODEL_IDENTIFIER)
+model = SiglipForImageClassification.from_pretrained(MODEL_IDENTIFIER).to(DEVICE)
+model.eval()
+
+# FastAPI app
+app = FastAPI()
+
+@app.get("/")
+def root():
+    return {"message": "AI vs Human image detector is running."}
+
+@app.post("/predict")
+async def predict(file: UploadFile = File(...)):
+    try:
+        image_bytes = await file.read()
+        image = PILImage.open(io.BytesIO(image_bytes)).convert("RGB")
+
+        inputs = processor(images=image, return_tensors="pt").to(DEVICE)
+        with torch.no_grad():
+            outputs = model(**inputs)
+            probs = torch.softmax(outputs.logits, dim=-1)[0]
+            results = {
+                model.config.id2label[i]: round(prob.item(), 4)
+                for i, prob in enumerate(probs)
+            }
+        return JSONResponse(content={"prediction": results})
+    except Exception as e:
+        return JSONResponse(content={"error": str(e)}, status_code=500)