app.py

# app.py
import gradio as gr
import torch
from torchvision import transforms, models
import cv2
import numpy as np
from PIL import Image
from ultralytics import YOLO

def load_models():
    # Initialize YOLO
    yolo_model = YOLO('HockeyAI.pt')
    
    # Initialize SqueezeNet
    squeezenet_model = models.squeezenet1_1(weights=None)
    squeezenet_model.classifier[1] = torch.nn.Conv2d(512, 8, kernel_size=1)
    squeezenet_model.num_classes = 8
    squeezenet_model.load_state_dict(torch.load('best_model_squezenet.pth', map_location=torch.device('cpu')))
    squeezenet_model.eval()
    
    return yolo_model, squeezenet_model

def process_image(input_image):
    if input_image is None:
        return None
    
    # Convert to numpy array if needed
    if isinstance(input_image, str):
        image = cv2.imread(input_image)
        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    else:
        image = input_image.copy()
    
    # Initialize models
    yolo_model, squeezenet_model = load_models()
    
    # Class labels for direction
    class_labels = [
        "Bottom", "Bottom_Left", "Bottom_Right", "Left", 
        "Right", "Top", "Top_Left", "Top_Right"
    ]
    
    # Transform for SqueezeNet
    transform = transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
    ])
    
    # Run YOLO detection
    results = yolo_model(image)
    
    # Process each detection
    for box in results[0].boxes:
        xyxy = box.xyxy[0].cpu().numpy()
        conf = float(box.conf[0].cpu().numpy())
        cls = int(box.cls[0].cpu().numpy())
        
        # Process only if it's a player (class 4) and confidence is above threshold
        if cls == 4 and conf > 0.5:
            x1, y1, x2, y2 = map(int, xyxy)
            
            # Crop and process for direction classification
            if x2 > x1 and y2 > y1:
                cropped_array = image[y1:y2, x1:x2]
                if cropped_array.size > 0:
                    cropped_image = Image.fromarray(cropped_array)
                    
                    # Predict direction
                    image_tensor = transform(cropped_image).unsqueeze(0)
                    with torch.no_grad():
                        output = squeezenet_model(image_tensor)
                        direction_class = torch.argmax(output, dim=1).item()
                    
                    # Draw annotations
                    cv2.rectangle(image, (x1, y1), (x2, y2), (0, 255, 0), 2)
                    cv2.putText(image, f"{conf:.2f}", (x1, y1-10), 
                              cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
                    
                    # Draw direction arrow
                    center_x, center_y = (x1 + x2) // 2, (y1 + y2) // 2
                    arrow_length = 80  # Increased from 50 to 80
                    direction = class_labels[direction_class]
                    
                    # Calculate arrow endpoint
                    end_x, end_y = center_x, center_y
                    if "Top" in direction:
                        end_y = center_y - arrow_length
                    elif "Bottom" in direction:
                        end_y = center_y + arrow_length
                    if "Left" in direction:
                        end_x = center_x - arrow_length
                    elif "Right" in direction:
                        end_x = center_x + arrow_length
                        
                    cv2.arrowedLine(image, (center_x, center_y), (end_x, end_y), 
                                  (255, 0, 0), 4, tipLength=0.4)  
    
    return image

# Create Gradio interface
def gradio_interface():
    with gr.Blocks() as iface:
        gr.Markdown("# Player Direction Detection")
        gr.Markdown("Upload an image to detect players and their movement directions")
        
        with gr.Row():
            with gr.Column():
                input_image = gr.Image(label="Input Image", type="numpy")
            with gr.Column():
                output_image = gr.Image(label="Output Image")
        
        # Handle image processing
        input_image.change(
            fn=process_image,
            inputs=[input_image],
            outputs=[output_image]
        )
        
        # Add example images if you have them
        gr.Examples(
            examples=["example-1.jpg", "example-2.jpg"],
            inputs=input_image,
            outputs=output_image,
            fn=process_image,
            cache_examples=True
        )
    
    return iface

if __name__ == "__main__":
    iface = gradio_interface()
    iface.launch()