pyav15.0 reuses input stream as template, time base is not copied to output stream

[ncheng89]

Reusing the input stream as a template for the output stream does not seem to copy the time base of the input stream to the output stream as a base

import av

INPUT_URL = "rtsp://abc:abc@192.168.11.134:554/Streaming/Channels/101"

in_container = av.open(INPUT_URL, options={"rtsp_transport": "tcp", "timeout": "5000000"})
video_stream = in_container.streams.video[0]

print("Input stream time_base:", video_stream.time_base)

out_container = av.open("test_output.mp4", mode="w")

out_stream = out_container.add_stream_from_template(video_stream)

print("Output stream time_base (default):", out_stream.time_base)

out_stream.time_base = video_stream.time_base
print("Output stream time_base (after fix):", out_stream.time_base)

out_container.close()
in_container.close()

Output log

Input stream time_base: 1/90000
Output stream time_base (default): None
Output stream time_base (after fix): 1/90000

[abdullayevziyodulla]

This is version-dependent: the current add_stream_from_template implementation explicitly copies the template's stream time base into both the new stream and its codec context. That differs from the None observed in your PyAV 15 example.

For a release where it remains unset, assigning it before writing, as you did, is a reasonable explicit initialization. However, the muxer can choose a different stream time base when the output header is written. Inspect the value after out_container.start_encoding() if you need the finalized scale.

For remuxing, keep each input packet's existing timestamp scale, assign its output stream, and let mux() rescale the packet to that stream's time base:

out_stream = out_container.add_stream_from_template(video_stream)
out_stream.time_base = video_stream.time_base
for packet in in_container.demux(video_stream):
    if packet.dts is None:
        continue
    packet.stream = out_stream
    out_container.mux(packet)

Don't simply label existing numeric PTS/DTS with a new time base: that would change their meaning. Current mux_one calls packet._rebase_time(stream.time_base) internally. The timing documentation explains why input, codec and output stream scales need to be distinguished.