[FFmpeg-devel] [PATCH] avfilter/vf_xcam: add xcam video filter

[Zong, Wei]

> -----Original Message-----
> From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of Paul B
> Mahol
> Sent: Friday, July 31, 2020 3:57 PM
> To: FFmpeg development discussions and patches <ffmpeg-devel at ffmpeg.org>
> Cc: Liu, YinhangX <yinhangx.liu at intel.com>
> Subject: Re: [FFmpeg-devel] [PATCH] avfilter/vf_xcam: add xcam video filter
> 
> On 7/31/20, Zong, Wei <wei.zong at intel.com> wrote:
> > Thank you for the comments on this patch, I want to make some
> > explanation on your comments.
> >
> > 1. the purpose of this patch
> >      We got some requirements from users who are trying libxcam
> > stitching or developers who showing interests. Many of them want an
> > easy way to construct an end 2 end pipeline of high quality immersive media
> application.
> >     The key points are high resolution and low latency, FFMpeg
> > provides high quality video codec framework, if we implement video
> > stitching as a video filter in FFMpeg, they will get benefits.
> >
> > 2. 3rd party dependency
> >     The functions exposed by xcam filter only depends on essential
> > build tool-chain.
> > According to hardware configuration, user can get performance
> > improvement by install OCL, GLES or Vulkan.
> > These libraries are not necessary for video stitching, user can build
> > a CPU version without these libraries.
> >     Beside these, stitching and DVS function depend on feature match
> > algorithm from OpenCV library to get better quality, I posted demo
> > link in commit message.
> > User can run stitching without install OpenCV, the side effect is the
> > quality will decrease.
> >
> > 3. Immersive media standard
> >     Normally the camera manufactures provide video stitching software,
> > they defined private camera calibration parameters, these are very
> > critical to stitching quality. I know MPEG is making proposal on
> > transfer these meta-data in video stream, this will help compose
> > immersive media on cloud or edge server. I think it's a good idea to
> > create such a video filter in FFMpeg pipeline.
> >
> > 4. Why not implement stitching inside FFMpeg video filter
> >     The algorithms are complexity and the performance are also
> > important, the stitching function manage computing resource for
> > example thread handling, image data block dispatching. Personally I
> > think we don't need to construct a wheel from the beginning.
> 
> Number of supported pixel formats are rather very low.

   Do you mean the pixel format (NV12 & YUV420) that xcam filter supported is not enough?
 xcam filter works as a post processing component between video decoder and encoder, we suppose the data comes from camera or stored video clips, so it handles YUV format.
libxcam algorithm optimization highly depends on pixel layout, to support YUV422 or YUV444 will cost much effort, we do not meet the requirement till now.


> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email ffmpeg-devel-request at ffmpeg.org
> with subject "unsubscribe".