On Tue, Feb 18, 2020 at 11:06 AM Guo, Yejun <yejun....@intel.com> wrote: > > > > > -----Original Message----- > > From: Guo, Yejun > > Sent: Tuesday, February 18, 2020 10:35 AM > > To: Guo, Yejun <yejun....@intel.com> > > Subject: RE: [FFmpeg-devel] [PATCH 1/3] avutil/frame: add use_last_roi > > > > > > > > From: Ronald S. Bultje [mailto:rsbul...@gmail.com] > > Sent: Friday, February 14, 2020 11:02 PM > > To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org> > > Cc: Guo, Yejun <yejun....@intel.com> > > Subject: Re: [FFmpeg-devel] [PATCH 1/3] avutil/frame: add use_last_roi > > > > Hi, > > > >> On Thu, Feb 13, 2020 at 10:41 PM Guo, Yejun <yejun....@intel.com> wrote: > >> For some cases, the regions of interest do not change, it is not > >> convenient to always prepare the roi data for every frame. > > > > Since side-data is refcounted, can't you just keep the "last" one in memory > > and > > refcount it? That way, if you wanted to know if it's the same, you can > > compare > > pointers, or if you want to access the ROI data directly, you can do that, > > but it > > doesn't require a copy. > > > > Ronald > > Thanks Ronald, yes, the application developers can 'cache' the last > 'AVBufferRef *' of > the roi side-data, so they don't need to prepare the same side-data for every > frame, just > pass 'AVBufferRef*' to side-data. So, we don't need to add use_last_roi in > AVFrame. > > The only difference is that we need to re-calculate rois for encoders within > ffmpeg, we > can optimize it with your method if it is found to be a bottle neck. I don't think re-calculate rois per frame is a true bottle neck, but have a method to avoid unnecessary re-calculate is better. _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
