Add app.py

Author: farhann-saleem

projects/detection/DeepCrop/src/app.py

@@ -0,0 +1,76 @@
+from flask import Flask, request, jsonify
+from flask_cors import CORS
+from tensorflow.keras.models import load_model
+from tensorflow.keras.preprocessing import image
+import numpy as np
+import os
+import uuid
+
+app = Flask(__name__)
+CORS(app)  # allow React frontend to talk to Flask backend
+
+# ---- config ----
+MODEL_PATH = "../models/model.keras"           # your model file
+IMG_SIZE = (128, 128)                # must match your training size
+UPLOAD_DIR = "../data/temp"                  # temporary save for incoming files
+os.makedirs(UPLOAD_DIR, exist_ok=True)
+
+# These are the 3 classes we want to serve in the app UI
+SERVE_LABELS = ["Healthy", "Early Blight", "Late Blight"]
+
+# ---- load model once ----
+model = load_model(MODEL_PATH)
+
+
+@app.route("/predict", methods=["POST"])
+def predict():
+    if "file" not in request.files or request.files["file"].filename == "":
+        return jsonify({"error": "No file uploaded"}), 400
+
+    f = request.files["file"]
+
+    # save with random name to avoid collisions on repeated tests
+    ext = os.path.splitext(f.filename)[1].lower()
+    fname = f"{uuid.uuid4().hex}{ext if ext else '.jpg'}"
+    fpath = os.path.join(UPLOAD_DIR, fname)
+    f.save(fpath)
+
+    # --- preprocess exactly like training ---
+    img = image.load_img(fpath, target_size=IMG_SIZE)
+    arr = image.img_to_array(img)
+    arr = np.expand_dims(arr, 0) / 255.0
+
+    # --- raw prediction (model outputs 10 classes in your case) ---
+    preds = model.predict(arr)
+
+    # ---- IMPORTANT: keep only the first 3 outputs (Healthy, Early, Late)
+    # If your model’s class order differs, adjust the slice or indices here.
+    trimmed = preds[0][:3].astype(np.float64)
+
+    # renormalize the 3 scores so they sum to 1
+    total = np.sum(trimmed)
+    if total > 0:
+        trimmed /= total
+
+    idx = int(np.argmax(trimmed))
+    conf = float(trimmed[idx])
+
+    # logging for debugging in console
+    print("Raw preds:", preds)
+    print("Trimmed (3-class) preds:", trimmed)
+    print("Predicted:", SERVE_LABELS[idx], "Confidence:", conf)
+
+    return jsonify({
+        "label": SERVE_LABELS[idx],
+        "confidence": round(conf * 100, 2)
+    })
+
+
+@app.route("/", methods=["GET"])
+def health():
+    # simple health check
+    return jsonify({"ok": True, "model": MODEL_PATH})
+
+
+if __name__ == "__main__":
+    app.run(port=5000, debug=True)