libvlc ffmpeg：在mpegts h264流中不查找

Question

I am using ffmpeg to record video input from GDI (windows screen recorder) to view it later using VLC (via ActiveX plugin) + ffmpeg to decode it.

Right now seeking in video is not working in VLC via plugin (which is critical). VLC player itself provide seeking, but it is more like byte position seeking (on I- frames which are larger than other frames it makes larger steps on horizontal scroll and also there are no timestamps).

Encoder is opened with next defaults:

avformat_alloc_output_context2(&outputContext, NULL, "mpegts", "test.mpg");
outputFormat = outputContext->oformat;
encoder = avcodec_find_encoder(AV_CODEC_ID_H264);
outputStream = avformat_new_stream(outputContext, encoder);
outputStream->id = outputContext->nb_streams - 1;
encoderContext = outputStream->codec;
encoderContext->bit_rate = bitrate;   // 800000 by default
encoderContext->rc_max_rate = bitrate;
encoderContext->width = imageWidth;   // 1920
encoderContext->height = imageHeight; // 1080
encoderContext->time_base.num = 1;
encoderContext->time_base.den = fps;  // 25 by default
encoderContext->gop_size = fps;
encoderContext->keyint_min = fps;
encoderContext->max_b_frames = 0;
encoderContext->pix_fmt = AV_PIX_FMT_YUV420P;
outputStream->time_base = encoderContext->time_base;
avcodec_open2(encoderContext, encoder, NULL);

Recording is done this way:

// my impl of GDI recorder, returning AVFrame with only data and linesize filled.
AVFrame* tmp_frame = impl_->recorder->acquireFrame();

// converting RGB -> YUV420
sws_scale(impl_->scaleContext, tmp_frame->data, tmp_frame->linesize, 0, impl_->frame->height, impl_->frame->data, impl_->frame->linesize);

// pts variable is calculated by using QueryPerformanceCounter form WinAPI. It is strictly increasing
impl_->frame->pts = pts;

avcodec_encode_video2(impl_->encoderContext, impl_->packet, impl_->frame, &out_size);

if (out_size) {
     impl_->packet->pts = pts;
     impl_->packet->dts = pts;
     impl_->packet->duration = 1; // here it is! It is set but has no effect
     av_packet_rescale_ts(impl_->packet, impl_->encoderContext->time_base, impl_->outputStream->time_base);
     // here pts = 3600*pts, dts = 3600*pts, duration = 3600 what I consider to be legit in terms of milliseconds
     impl_->packet->stream_index = impl_->outputStream->index;
     av_interleaved_write_frame(impl_->outputContext, impl_->packet);
     av_packet_unref(impl_->packet);
     out_size = 0;
}

ffprobe is providing next info on frames:

[FRAME]
media_type=video
stream_index=0
key_frame=1
pkt_pts=3600
pkt_pts_time=0:00:00.040000
pkt_dts=3600
pkt_dts_time=0:00:00.040000
best_effort_timestamp=3600
best_effort_timestamp_time=0:00:00.040000
pkt_duration=N/A
pkt_duration_time=N/A
pkt_pos=564
pkt_size=97.018555 Kibyte
width=1920
height=1080
pix_fmt=yuv420p
sample_aspect_ratio=N/A
pict_type=I
coded_picture_number=0
display_picture_number=0
interlaced_frame=0
top_field_first=0
repeat_pict=0
[/FRAME]

I believe that problem is in pkt_duration variable, though it was set. What I am doing wrong in recording so I can't seek in video?

PS on other videos (also h264) seeking is working in ActiveX VLC plugin.

Answer 1

What is definitely wrong, is:

impl_->packet->pts = pts;
impl_->packet->dts = pts;

PTS and DTS are not equal! They could be if you would have only I-frames, which is not the case here. Also, your comment says: pts variable is calculated by using QueryPerformanceCounter form WinAPI . If your frame rate is constant, and I believe it is, then you don't need QueryPerformanceCounter API. PTS is usually in 90kHz units. The duration of 1 frame expressed in 90kHz is calculated like this:

90000 x denominator / numerator

If fps is 25 then numerator is 25 and denominator is 1. For 29.97 fps the numerator is 30000 and denominator is 1001. Each new frame's PTS should be increased for that amount (unless you have dropped frames). Regarding the DTS, the encoder should provide that value.