Andyson's picture
Upload videos_zip_two_refiners/scripts/concat_video_chunks.py with huggingface_hub
d263724 verified
Raw
History Blame Contribute Delete
2.67 kB
#!/usr/bin/env python
"""Concatenate MP4 chunks by decoding and re-encoding with PyAV."""
from __future__ import annotations
import argparse
import json
from fractions import Fraction
from pathlib import Path
import av
def concat_videos(inputs: list[Path], output_path: Path, fps: float) -> int:
if not inputs:
raise ValueError("No input chunks to concatenate")
first = av.open(str(inputs[0]))
try:
first_stream = first.streams.video[0]
width = first_stream.width
height = first_stream.height
finally:
first.close()
output_path.parent.mkdir(parents=True, exist_ok=True)
container = av.open(str(output_path), "w")
success = False
frame_count = 0
try:
output_fps = round(fps)
time_base = Fraction(1, output_fps)
stream = container.add_stream("libx264", rate=output_fps, options={"crf": "19", "preset": "veryfast"})
stream.width = width
stream.height = height
stream.pix_fmt = "yuv420p"
stream.time_base = time_base
stream.codec_context.thread_count = 0
stream.codec_context.thread_type = "FRAME"
for input_path in inputs:
chunk = av.open(str(input_path))
try:
video_stream = chunk.streams.video[0]
if video_stream.width != width or video_stream.height != height:
raise ValueError(
f"Chunk {input_path} is {video_stream.width}x{video_stream.height}, "
f"expected {width}x{height}"
)
for frame in chunk.decode(video_stream):
frame = frame.reformat(width=width, height=height, format="yuv420p")
frame.pts = frame_count
frame.time_base = time_base
for packet in stream.encode(frame):
container.mux(packet)
frame_count += 1
finally:
chunk.close()
for packet in stream.encode():
container.mux(packet)
success = True
finally:
container.close()
if not success:
output_path.unlink(missing_ok=True)
return frame_count
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--output", required=True)
parser.add_argument("--fps", type=float, required=True)
parser.add_argument("inputs", nargs="+")
args = parser.parse_args()
frames = concat_videos([Path(p) for p in args.inputs], Path(args.output), args.fps)
print(json.dumps({"frames": frames}))
if __name__ == "__main__":
main()