[FFmpeg-devel] [PATCH] examples/transcoding: Fix time_base handling

[Jack Lau]

Hello.

I found one better method to fix this issue. Firstly, i think the problem
caused by incorrectly setting enc_ctx->time_base, so i refer to the
implementation in avcodec_open2()

361 #if FF_API_AVCTX_TIMEBASE
362        if (avctx->framerate.num > 0 && avctx->framerate.den > 0)
363             avctx->time_base = av_inv_q(av_mul_q(avctx->framerate,
(AVRational){avctx->ticks_per_frame, 1}));
364 #endif

this is new version code and i submit it in new mail, it can be reviewed in
https://patchwork.ffmpeg.org/project/ffmpeg/patch/tencent_7ADFE69819D49D96FA8BC2C9E688CCEC2506@qq.com/
-                enc_ctx->time_base = av_inv_q(dec_ctx->framerate);
+                enc_ctx->time_base = av_inv_q(av_mul_q(dec_ctx->framerate,
(AVRational){dec_ctx->ticks_per_frame, 1}));

-                enc_ctx->time_base = (AVRational){1, enc_ctx->sample_rate};
+                enc_ctx->time_base = (AVRational){1, dec_ctx->sample_rate};

but there is new issue in andriy/configure_x86 check:

error: Failed to merge in the changes.

i think it's maybe the mistake of branch? this patch is for release/5.1
branch rather than master branch.

Could any one give me some advice please so that i can fix it?


On Wed, Feb 5, 2025 at 11:25 AM Jack Lau <jacklau1222gm at gmail.com> wrote:

> To be clear, i want to give an example, i use a 10s duration, 30fps video.
> The ifmt_ctx->streams[stream_index]->time_base is same as
> ofmt_ctx->streams[stream_index]->time_base after initializing.
> The stream->dec_ctx->time_base is default 0,60 but
> stream->enc_ctx->time_base is set to av_inv_q(dec_ctx->framerate) so it's
> 0,30
>
> 174                 /* video time_base can be set to whatever is handy and
> supported by encoder */
> 175                 enc_ctx->time_base = av_inv_q(dec_ctx->framerate);
>
> so the twice rescale is this:
> input pkt in_tb: 15360   out_tb: 60
> output pkt in_tb: 30     out_tb: 15360
>
> so i get one 20s duration and 15fps video(audio duration is normal because
> the input's audio sample ratio is 44100)
>
> So i think the problem is that the enc_ctx->time_base shouldn't set to
> av_inv_q(dec_ctx->framerate)
>
> On Wed, Feb 5, 2025 at 10:29 AM Jack Lau <jacklau1222gm at gmail.com> wrote:
>
>>
>> AVCodecContext.time_base is not used for decoding.
>>
>>
>>
>> Thank you for your reply. I understand that time_base is not used during
>> decoding, but the transcoding code calls av_packet_rescale_ts twice,
>> once before decoding and once after encoding, as shown below:
>>
>> 540         if (filter_ctx[stream_index].filter_graph) {
>> 541             StreamContext *stream = &stream_ctx[stream_index];
>> 542
>> 543             av_log(NULL, AV_LOG_DEBUG, "Going to reencode&filter the
>> frame\n");
>> 544
>> 545             av_packet_rescale_ts(packet,
>>
>>
>> 546
>>  ifmt_ctx->streams[stream_index]->time_base,
>> 547                                  stream->dec_ctx->time_base);
>> 548             ret = avcodec_send_packet(stream->dec_ctx, packet);
>>
>> 448         /* prepare packet for muxing */
>>
>>
>> 449         enc_pkt->stream_index = stream_index;
>> 450         av_packet_rescale_ts(enc_pkt,
>> 451                              stream->enc_ctx->time_base,
>> 452
>>  ofmt_ctx->streams[stream_index]->time_base);
>> 453
>> 454         av_log(NULL, AV_LOG_DEBUG, "Muxing frame\n");
>> 455         /* mux encoded frame */
>> 456         ret = av_interleaved_write_frame(ofmt_ctx, enc_pkt);
>>
>> If dec_ctx->time_base and enc_ctx->time_base are not consistent, it can
>> lead to incorrect packet duration, which in turn causes issues with the
>> output file's duration.
>>
>