[x264-devel] Re: Questions about x264 encoder apis

[Bill May]

Loren and Måns,

Thanks for your quick responses.

Loren Merritt wrote:
> We do not modify the input data. Yes, such an optimization would be 
> possible. Note that there may currently be some assumptions about the 
> strides that would have to be met or fixed.

Great.  Thank you.  Another question - is it possible to have access to
the reconstructed YUV from the encoder, so we can display a preview ?  If
you tell me which structure, I'd be glad to write an api and contribute it
back.

>> 2). Do you have any plans to install the library and include files so 
>> they could be detected by other projects via configure ?
> 
> 
> Done. Unless you mean some more complicated automake system, in which 
> case, no I have no such plans.

And likewise, perfect.  Now I can detect the library.

>> 5). i_bitrate is in bits per second ?  If I use b_cbr, what should I 
>> set the other parameters to ?
> 
> 
> kbit/sec.
> Good values are rc_buffer_size = .5-1 seconds worth of bits (measured in 
> kbit), and rc_init_buffer = .5*rc_buffer_size (measured in kbit). Or 
> just leave them 0 and let x264 fill them in based on bitrate.

I'll leave it 0.

>> 6). Any ideas about which nerd knobs would make sense (b_cabac, etc) ?
> 
> 
> cabac should always be on, no need to expose it.
> The main options (regarding quality vs speed) are subpel_refine, 
> frame_reference, and bframe.
> For most of the other options: you can play with them if you know what 
> you're doing, but otherwise they're just there for my own convenience in 
> searching for good defaults.

Thanks.  I'm going to try b-frames soon, but wanted to get the base release
out for people to test.

BTW - I also wanted to commend you for putting the pts of the encoded frame
in the x264_picture_t.

Thanks very much for your prompt responses.  You guys rock. (Sorry to sound
like a fanboy, but this has been the easiest integration I've done in 4
years of working on mpeg4ip).

Bill

-- 
This is the x264-devel mailing-list
To unsubscribe, go to: http://developers.videolan.org/lists.html