Update app.py

app.py

@@ -1,11 +1,28 @@
-import gradio as gr
+import os, sys, subprocess, torch, time, cv2, numpy as np, gradio as gr
+
+# ==============================================================
+# 🔒 Force correct Ultralytics version (prevents HF auto-upgrade)
+# ==============================================================
+subprocess.run(
+    [sys.executable, "-m", "pip", "uninstall", "-y", "ultralytics"],
+    stdout=subprocess.DEVNULL,
+    stderr=subprocess.DEVNULL,
+)
+subprocess.run(
+    [
+        sys.executable, "-m", "pip", "install",
+        "--no-deps", "--force-reinstall",
+        "git+https://github.com/ultralytics/ultralytics.git@v8.0.20"
+    ],
+    stdout=subprocess.DEVNULL,
+    stderr=subprocess.DEVNULL,
+)
+
+# ==============================================================
+# 🚀 Import YOLO safely (after version is locked)
+# ==============================================================
 from ultralytics import YOLO
-import cv2
-import numpy as np
-import torch
-import time
 
-# 🚀 Load model
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
 print(f"🚀 Using device: {device}")
 
@@ -15,34 +32,32 @@ model.to(device)
 print("✅ Model loaded successfully!")
 print("Model class names:", model.names)
 
+
+# ==============================================================
+# 🧠 Inference Function
+# ==============================================================
 def detect(image):
     """
-    image: numpy array (Gradio automatically provides this)
-    returns: annotated image and detected conclusion text
+    Run YOLO inference on the uploaded image and classify tomato quality.
     """
     try:
         start_time = time.time()
         
-        # 🧠 YOLO inference
+        # Run YOLO prediction
         results = model.predict(
-            image,
+            source=image,
             imgsz=640,
             conf=0.25,
             iou=0.45,
             augment=True,
-            verbose=False
+            verbose=False,
+            device=device
         )
 
         annotated = results[0].plot()
+        detected_classes = [model.names[int(box.cls[0])].strip().lower() for box in results[0].boxes]
 
-        detected_classes = []
-        for box in results[0].boxes:
-            cls = int(box.cls[0])
-            conf = float(box.conf[0])
-            label = model.names[cls].strip().lower()
-            detected_classes.append(label)
-
-        # ✅ Decision logic
+        # Quality classification logic
         conclusion = "No Tomato Detected"
         if any("damaged" in c for c in detected_classes):
             conclusion = "Damaged 🍂"
@@ -60,7 +75,10 @@ def detect(image):
         print("❌ Error:", e)
         return image, f"Error: {str(e)}"
 
-# 🧩 Build Gradio Interface
+
+# ==============================================================
+# 🖥️ Gradio Interface
+# ==============================================================
 interface = gr.Interface(
     fn=detect,
     inputs=gr.Image(type="numpy", label="Upload Tomato Image"),
@@ -70,7 +88,6 @@ interface = gr.Interface(
     ],
     title="Tomato Quality Detector 🍅",
     description="Upload a tomato image to detect its quality (Ripe / Unripe / Damaged) using a YOLOv8 model fine-tuned on a custom dataset.",
-    examples=None,
     theme="default"
 )