evo-video-sampling

Extracts frames from an MP4 video at a target FPS using reliable sequential reading (avoids unreliable CAP_PROP_POS_FRAMES seeking in compressed MP4s).

Key Function

• sample_video_frames(video_path, target_fps=6) — Returns (sampled_frames, frame_indices, original_fps)
  • sampled_frames: list of grayscale np.uint8 2D arrays
  • frame_indices: list of original frame indices (0-based)
  • original_fps: the video's native FPS

Usage

import sys
sys.path.insert(0, '/app/environment/skills/evo-video-sampling/scripts')
from utils import sample_video_frames

frames, indices, fps = sample_video_frames('/root/input.mp4', target_fps=6)
print(f"Sampled {len(frames)} frames from video at {fps} FPS")

Implementation Details

• Uses cv2.VideoCapture with sequential read() loop (no seeking)
• Frame step = max(1, int(round(original_fps / target_fps)))
• Converts BGR to grayscale immediately via cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
• Fallback FPS of 30.0 if CAP_PROP_FPS returns <= 0