[x265] Adaptove GoP sizes - not sure if its working

[Steve Borho]

On 10/07, Xinyue Lu wrote:
> Hi,
> 
> Thanks for the test result.
> 
> To test the previous behavior, pick a random version on dev branch
> before r10850 should be good enough.
> 
> I haven't had some time to test through it, but, if with the latest
> patch, it still produces I frames at 0, 250, 500 on my test video
> clips, then that's unacceptable for me. There *are* real scene cuts,
> and the encoder should prefer inserting I frames at scene cuts than
> the max gop length.

this blanket statement is incorrect. If the scene cut happens a large
distance from the max keyframe interval it is better to encode it as a P
to save bits. When you specify --keyint-max 250, the encoder will try to
only use one I frame per 250 frames, this is its goal.

If there are scene cuts near the max keyframe interval that are not
being set to I, then that could be a problem. The bias algorithm does
tend to detect many possible scene cuts near the end of the max interval
and this tends to trigger the flash detection logic (where you have many
scene cut frames in a row) which defers the I frame until the end of the
flash or max interval (see scenecut.cpp line 1450) It's not possible to
tell if this is happening in your case without seeing the input video.

-- 
Steve Borho