Performance Considerations

import cv2
import numpy as np
import time

def benchmark_operations(video_path, n_frames=100):
    """Benchmark different operations to identify bottlenecks."""
    cap = cv2.VideoCapture(video_path)
    times = {}

    frames = []
    for _ in range(n_frames):
        ret, frame = cap.read()
        if ret:
            frames.append(frame)
    cap.release()

    if not frames:
        print("No frames loaded")
        return

    print(f"Benchmarking on {len(frames)} frames...")

    # Benchmark grayscale
    t0 = time.time()
    for f in frames:
        cv2.cvtColor(f, cv2.COLOR_BGR2GRAY)
    times["grayscale"] = (time.time() - t0) / len(frames) * 1000

    # Benchmark resize
    t0 = time.time()
    for f in frames:
        cv2.resize(f, (320, 240))
    times["resize_320x240"] = (time.time() - t0) / len(frames) * 1000

    # Benchmark Gaussian blur
    t0 = time.time()
    for f in frames:
        cv2.GaussianBlur(f, (5, 5), 0)
    times["gaussian_blur"] = (time.time() - t0) / len(frames) * 1000

    # Benchmark Canny
    grays = [cv2.cvtColor(f, cv2.COLOR_BGR2GRAY) for f in frames]
    t0 = time.time()
    for g in grays:
        cv2.Canny(g, 50, 150)
    times["canny_edges"] = (time.time() - t0) / len(frames) * 1000

    print("\nOperation times (ms per frame):")
    for op, ms in sorted(times.items(), key=lambda x: x[1]):
        fps_cap = 1000 / ms if ms > 0 else float("inf")
        print(f"  {op:20s}: {ms:.2f} ms → max {fps_cap:.0f} FPS")

# Tips for optimising:
print("\nOptimisation strategies:")
tips = [
    "Resize frames down before processing (640x480 instead of 1920x1080)",
    "Process only every Nth frame for non-real-time tasks",
    "Use grayscale when colour is not needed",
    "Use cv2.INTER_AREA when shrinking (faster than INTER_CUBIC)",
    "Enable GPU acceleration: net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)",
    "Profile first — optimise only the bottleneck operation",
]
for tip in tips:
    print(f"  • {tip}")