Update app.py

app.py

@@ -3,10 +3,17 @@ import torch
 import cv2
 import numpy as np
 import matplotlib.pyplot as plt
-from yolov5 import YOLOv5
+from pathlib import Path
 
-# Load the trained YOLOv5 model (best.pt) using the YOLOv5 library
-model = YOLOv5("best.pt")  # This will use YOLOv5's internal loading mechanism
+# Load the trained YOLOv5 model from a local path (directly using PyTorch)
+def load_model(model_path):
+    # Load model using torch.load, ensuring no downloading from Hugging Face
+    model = torch.load(model_path, map_location="cpu")  # Load model from the local path
+    model.eval()  # Set the model to evaluation mode
+    return model
+
+# Load the model directly from the local file
+model = load_model("best.pt")  # Ensure 'best.pt' is in the correct path
 
 # Function to process the video and calculate ball trajectory, speed, and visualize the pitch
 def process_video(video_file):
@@ -25,12 +32,12 @@ def process_video(video_file):
         
         frame_count += 1
 
-        # Run YOLOv5 model on the frame to detect ball
+        # Run the model on the frame to detect the ball
         results = model(frame)  # Using the pre-trained YOLOv5 model
 
         # Extract the ball position (assuming class 0 = ball)
         ball_detections = results.pandas().xywh
-        ball = ball_detections[ball_detections['class'] == 0]  # class 0 is ball, adjust as needed
+        ball = ball_detections[ball_detections['class'] == 0]  # Class 0 for ball (adjust if needed)
         
         if not ball.empty:
             ball_x = ball.iloc[0]['xmin'] + (ball.iloc[0]['xmax'] - ball.iloc[0]['xmin']) / 2