video_utils.py

"""Video processing — extract frames from uploaded videos."""

import logging
import numpy as np

log = logging.getLogger("surprise.video")


def validate_video(path: str) -> dict:
    """Open the video and return metadata. Raises ValueError on failure."""
    import cv2

    cap = cv2.VideoCapture(path)
    if not cap.isOpened():
        raise ValueError("Could not open video file (unsupported format?)")

    fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
    n_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
    width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
    height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
    duration = n_frames / fps if fps > 0 else 0
    cap.release()

    if n_frames < 4:
        raise ValueError(f"Video too short: only {n_frames} frames (need ≥4)")

    return {
        "fps": float(fps),
        "n_frames": n_frames,
        "width": width,
        "height": height,
        "duration": float(duration),
    }


def extract_frames(
    path: str,
    target_size: int = 224,
    max_frames: int = 240,
    target_fps: int = 8,
) -> np.ndarray:
    """
    Extract frames from video, resampled to target_fps and resized to target_size.

    Returns:
        np.ndarray of shape (N, H, W, 3), dtype uint8, RGB.
    """
    import cv2

    cap = cv2.VideoCapture(path)
    if not cap.isOpened():
        raise ValueError("Cannot open video")

    src_fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
    stride = max(1, int(round(src_fps / target_fps)))

    frames = []
    idx = 0
    while True:
        ret, frame = cap.read()
        if not ret:
            break
        if idx % stride == 0:
            frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
            frame = _center_crop_resize(frame, target_size)
            frames.append(frame)
            if len(frames) >= max_frames:
                break
        idx += 1

    cap.release()
    if len(frames) < 4:
        raise ValueError(f"Too few frames extracted ({len(frames)})")
    return np.stack(frames, axis=0)


def _center_crop_resize(img: np.ndarray, size: int) -> np.ndarray:
    """Center-crop to square then resize to (size, size)."""
    import cv2

    h, w = img.shape[:2]
    s = min(h, w)
    y0 = (h - s) // 2
    x0 = (w - s) // 2
    img = img[y0 : y0 + s, x0 : x0 + s]
    img = cv2.resize(img, (size, size), interpolation=cv2.INTER_AREA)
    return img