Frame timestamps not preserved as-is during encoding

[oakaigh]

PyAV v17.0.1

Code to reproduce:

from fractions import Fraction
import av

path = "/tmp/pyav_nonzero_video_pts_repro.mp4"

fps = Fraction(30, 1)


out = av.open(path, "w")

stream = out.add_stream("mpeg4", rate=fps)
stream.width = 64
stream.height = 64
stream.pix_fmt = "yuv420p"

stream.codec_context.time_base = Fraction(1, fps)

print("encode input:")
for i in range(4):
    frame = av.VideoFrame(64, 64, "yuv420p")

    frame.pts = i + 1
    frame.time_base = source_tb

    print(
        i,
        "frame_time * fps =", Fraction(frame.pts) * frame.time_base * fps,
    )

    for packet in stream.encode(frame):
        out.mux(packet)

for packet in stream.encode():
    print(
        "  packet before mux:",
        "packet.pts =", packet.pts,
        "packet.time_base =", packet.time_base,
    )
    out.mux(packet)

out.close()

print("\ndecode output:")
inp = av.open(path)
video = inp.streams.video[0]

print(
    "decoded stream:",
    "time_base =", video.time_base,
    "start_time =", video.start_time,
    "average_rate =", video.average_rate,
)

for i, frame in enumerate(inp.decode(video)):
    t_x_fps = Fraction(frame.pts) * frame.time_base * fps

    print(
        i,
        "frame_time * fps =", t_x_fps,
    )

Output:

encode input:
0 frame_time * fps = 1
1 frame_time * fps = 2
2 frame_time * fps = 3
3 frame_time * fps = 4

decode output:
decoded stream: time_base = 1/15360 start_time = 507 average_rate = 30
0 frame_time * fps = 507/512
1 frame_time * fps = 1019/512
2 frame_time * fps = 1531/512
3 frame_time * fps = 2043/512

Expected output:

encode input:
0 frame_time * fps = 1
1 frame_time * fps = 2
2 frame_time * fps = 3
3 frame_time * fps = 4

decode output:
...
0 frame_time * fps = 1
1 frame_time * fps = 2
2 frame_time * fps = 3
3 frame_time * fps = 4

Weirdly enough if the pts starts from values like 0, 3, 6, 12, the problem does not occur at all.

[oakaigh]

Issue appears to be fixed after setting an appropriate movie timescale - document in examples?

from fractions import Fraction
import av

path = "/tmp/pyav_nonzero_video_pts_repro.mp4"

fps = Fraction(30, 1)


from math import lcm

movie_timescale = lcm(1000, fps.numerator)

out = av.open(path, "w",     container_options={
        "movie_timescale": str(movie_timescale),
    },)

stream = out.add_stream("mpeg4", rate=fps)
stream.width = 64
stream.height = 64
stream.pix_fmt = "yuv420p"

stream.codec_context.time_base = Fraction(1, fps)

print("encode input:")
for i in range(4):
    frame = av.VideoFrame(64, 64, "yuv420p")

    frame.pts = i + 2
    frame.time_base = source_tb

    print(
        i,
        # "frame.pts =", frame.pts,
        # "frame.time_base =", frame.time_base,
        "frame_time * fps =", Fraction(frame.pts) * frame.time_base * fps,
    )

    for packet in stream.encode(frame):
        # print(
        #     "  packet before mux:",
        #     "packet.pts =", packet.pts,
        #     "packet.time_base =", packet.time_base,
        # )
        out.mux(packet)

for packet in stream.encode():
    print(
        "  packet before mux:",
        "packet.pts =", packet.pts,
        "packet.time_base =", packet.time_base,
    )
    out.mux(packet)

out.close()

print("\ndecode output:")
inp = av.open(path)
video = inp.streams.video[0]

print(
    "decoded stream:",
    "time_base =", video.time_base,
    "start_time =", video.start_time,
    "average_rate =", video.average_rate,
)

for i, frame in enumerate(inp.decode(video)):
    t_x_fps = Fraction(frame.pts) * frame.time_base * fps

    print(
        i,
        # "frame.pts =", frame.pts,
        # "frame.time_base =", frame.time_base,
        "frame_time * fps =", t_x_fps,
        # "aligned =", t_x_fps.denominator == 1,
    )