app.py

import torch
import numpy as np
import gradio as gr
import cv2
import time
import os
from pathlib import Path
from PIL import Image

# Create cache directory for models
os.makedirs("models", exist_ok=True)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")

# Load YOLOv5 Nano model
model_path = Path("models/yolov5n.pt")
if model_path.exists():
    print(f"Loading model from cache: {model_path}")
    model = torch.hub.load("ultralytics/yolov5", "custom", path=str(model_path), source="local").to(device)
else:
    print("Downloading YOLOv5n model and caching...")
    model = torch.hub.load("ultralytics/yolov5", "yolov5n", pretrained=True).to(device)
    torch.save(model.state_dict(), model_path)

# Optimize model for speed
model.conf = 0.3  # Lower confidence threshold
model.iou = 0.3   # Non-Maximum Suppression IoU threshold
model.classes = None  # Detect all 80+ COCO classes

if device.type == "cuda":
    model.half()  # Use FP16 for faster inference
else:
    torch.set_num_threads(os.cpu_count())

model.eval()

# Pre-generate colors for bounding boxes
np.random.seed(42)
colors = np.random.randint(0, 255, size=(len(model.names), 3), dtype=np.uint8)

def process_video(video_path):
    cap = cv2.VideoCapture(video_path)
    
    if not cap.isOpened():
        return "Error: Could not open video file."

    frame_width = int(cap.get(3))
    frame_height = int(cap.get(4))
    fps = cap.get(cv2.CAP_PROP_FPS)
    
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    output_path = "output_video.mp4"
    out = cv2.VideoWriter(output_path, fourcc, fps, (frame_width, frame_height))

    total_frames = 0
    total_time = 0

    while cap.isOpened():
        ret, frame = cap.read()
        if not ret:
            break  # Break if no more frames
        
        start_time = time.time()
        
        # Convert frame for YOLOv5
        img = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
        results = model(img, size=640)
        
        inference_time = time.time() - start_time
        total_time += inference_time
        total_frames += 1
        
        detections = results.pred[0].cpu().numpy()

        for *xyxy, conf, cls in detections:
            x1, y1, x2, y2 = map(int, xyxy)
            class_id = int(cls)
            color = colors[class_id].tolist()
            cv2.rectangle(frame, (x1, y1), (x2, y2), color, 3, lineType=cv2.LINE_AA)
            label = f"{model.names[class_id]} {conf:.2f}"
            cv2.putText(frame, label, (x1, y1 - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (255, 255, 255), 2)

        # Calculate FPS
        avg_fps = total_frames / total_time if total_time > 0 else 0
        cv2.putText(frame, f"FPS: {avg_fps:.2f}", (20, 40), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)

        out.write(frame)

    cap.release()
    out.release()
    
    return output_path

def process_image(image):
    img = np.array(image)
    results = model(img, size=640)

    detections = results.pred[0].cpu().numpy()

    for *xyxy, conf, cls in detections:
        x1, y1, x2, y2 = map(int, xyxy)
        class_id = int(cls)
        color = colors[class_id].tolist()
        cv2.rectangle(img, (x1, y1), (x2, y2), color, 3, lineType=cv2.LINE_AA)
        label = f"{model.names[class_id]} {conf:.2f}"
        cv2.putText(img, label, (x1, y1 - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (255, 255, 255), 2)

    return Image.fromarray(img)

with gr.Blocks(title="Real-Time YOLOv5 Video & Image Object Detection") as demo:
    gr.Markdown("""
    # Real-Time YOLOv5 Object Detection
    """, elem_id="title")
    
    with gr.Tabs():
        with gr.TabItem("Video Detection"):
            with gr.Row():
                video_input = gr.Video(label="Upload Video", interactive=True, elem_id="video-input")
                process_button = gr.Button("Process Video", variant="primary", elem_id="video-process-btn")
            video_output = gr.Video(label="Processed Video", elem_id="video-output")
            process_button.click(fn=process_video, inputs=video_input, outputs=video_output)

        with gr.TabItem("Image Detection"):
            with gr.Row():
                image_input = gr.Image(type="pil", label="Upload Image", interactive=True)
            with gr.Row():
                clear_button = gr.Button("Clear", variant="secondary", elem_id="clear-btn")
                submit_button = gr.Button("Detect Objects", variant="primary", elem_id="submit-btn")
            clear_button.click(fn=lambda: None, inputs=None, outputs=image_output)
            submit_button.click(fn=process_image, inputs=image_input, outputs=image_output)
            with gr.Row():
                image_output = gr.Image(label="Detected Objects", elem_id="image-output")
            

demo.launch()