Update app.py

app.py

@@ -1,27 +1,26 @@
+from fastapi import FastAPI, File, UploadFile
+from fastapi.responses import JSONResponse
 import pickle
 import numpy as np
 from PIL import Image
-import gradio as gr
 
-# 1) Load your pretrained model
+app = FastAPI()
 model = pickle.load(open("mnist_model.pkl", "rb"))
 
-# 2) Define a prediction function
-def classify_digit(img):
-    # convert to grayscale 28×28
-    img = img.convert("L").resize((28, 28))
-    arr = np.array(img).reshape(1, -1)
-    pred = model.predict(arr)[0]
-    return f"Predicted digit: {pred}"
-
-# 3) Wire up Gradio
-iface = gr.Interface(
-    fn=classify_digit,
-    inputs=gr.Image(type="pil", label="Upload a 28×28 digit"),
-    outputs=gr.Textbox(label="Prediction"),
-    title="MNIST Digit Classifier",
-    description="Upload a handwritten digit and get a prediction!"
-)
+def preprocess_image(file_bytes) -> np.ndarray:
+    # 1) Load into PIL, convert to grayscale 'L'
+    img = Image.open(file_bytes).convert("L")
+    # 2) Resize to 28×28 (use ANTIALIAS for quality)
+    img = img.resize((28,28), Image.ANTIALIAS)
+    # 3) Convert to numpy array (uint8), flatten to length-784
+    arr = np.array(img).astype("uint8").reshape(1, -1)
+    # 4) Optionally invert colors if your MNIST is white-on-black:
+    #    arr = 255 - arr
+    return arr
 
-if __name__ == "__main__":
-    iface.launch()
+@app.post("/predict-image/")
+async def predict_image(file: UploadFile = File(...)):
+    # read the incoming UploadFile into BytesIO
+    arr = preprocess_image(file.file)
+    pred = model.predict(arr)[0]
+    return JSONResponse({"prediction": int(pred)})