Add seeking #14
Add seeking #14
Conversation
c_src/xav/reader.c
Outdated
|
|
||
| avcodec_flush_buffers(reader->c); | ||
|
|
||
| if (av_seek_frame(reader->fmt_ctx, reader->stream_idx, frmseekPos, AVSEEK_FLAG_BACKWARD) < 0) { |
There was a problem hiding this comment.
In my test video, there is a keyframe every second with a framerate of 25 fps. But when i seek, i can only get to frames which are a multiple of 250 frames...couldn't figure out where the issue is. Maybe we have to explicitly calculate P frames?
There was a problem hiding this comment.
That was indeed the issue! av_seek_frame does only seek to an I frame and you have to send packets to the decoder until you reach the desired timestamp. After some checking i confirmed that my test file had a keyframe every 10 seconds instead of every 1 seconds
c_src/xav/reader.c
Outdated
| int64_t frmseekPos = av_rescale_q( | ||
| (int64_t)(time_in_seconds * AV_TIME_BASE), | ||
| AV_TIME_BASE_Q, | ||
| reader->fmt_ctx->streams[reader->stream_idx]->time_base | ||
| ); |
There was a problem hiding this comment.
I think that what we need to do here is to convert time_in_seconds to frame number. Given that time_base is 1/fps we should do time_in_seconds * time_base.den / time_base.num so :
| int64_t frmseekPos = av_rescale_q( | |
| (int64_t)(time_in_seconds * AV_TIME_BASE), | |
| AV_TIME_BASE_Q, | |
| reader->fmt_ctx->streams[reader->stream_idx]->time_base | |
| ); | |
| av_rescale(time_in_seconds, time_base.den, time_base.num); |
There was a problem hiding this comment.
not exactly..av_seek_frame expects the seeking position in the same timebase as the selected stream not the frame number. I did some further research and it seems, that this is the correct way to calculate the seeking position and preserving the float accuracy. For example the file in test/fixtures/sample_h264.mp4 has a timebase of 1/15360 and AV_TIME_BASE is 1000000
There was a problem hiding this comment.
Ohh okay 🤦♂️ Could you please add a comment or link some resource that explains what we do here? I mean, what's the difference between AV_TIME_BASE and AVStream.time_base and why we need to use both in this calculation?
There was a problem hiding this comment.
Absolutely!
9ff14a7 to
89d7375
Compare
89d7375 to
59e850b
Compare
|
Perfect, thanks a lot! If we could just add a comment explaining what happens in timestamp calculation, we are ready to merge! |
45115dd to
79b93de
Compare
|
Merged, thank you! |
Hey, I saw this in you TODO issue #1 and could need it myself.
I've never written any C code, so I probably made some mistakes, but I hope this will help with some improvements.
I added a function to seek to a time in seconds, which will seek the
reader.Additionally, I've added the frame rate to the
Readerstruct, so the user can use this information for further calculations