[FFmpeg-devel] HAP QuickTime codec support

[bzk0711 at aol.com]

On Sat, 15 Nov 2014 12:34:57 +0100
Reimar Döffinger <Reimar.Doeffinger at gmx.de> wrote:

> > Are you sure this compression format makes any sense whatsoever?
> > You get a 1:6 compression (exactly, compared to RGB888), but
> > also horrible compression artefacts.
> > Simply going to half the resolution in x and y with YUV420 gives
> > you 1:8 compression and I would bet it would even look better.
> > At least I have a hard time seeing how it could look much worse.
> 
> Looking at texture compression formats, I don't think any make much
> sense for video that is originally in 4:2:0 format.
> The first that might be interesting (but is limited to a few mobile
> devices) is ASTC.
> Besides better quality at same bitrate, it allow for (relatively)
> fine-grained per-frame compression ratio selection (and thus a
> kind of rate-control), enough encoding options for perceptual encoding
> and temporal compression (by using 3D textures).
> But even with those improvements the usefulness for video even
> compared to uncompressed seems questionable, especially since even
> with uncompressed textures you could implement delta encoding
> on the GPU.

Well the intended use for me is solely to be able to
process more videos in parallel from a single hard
drive using the mentioned video compositing
application, quality is (for now) secondary, but I
will do some comparisons myself as soon as I can
get my hands on a MacOS or Windows machine again. 

The problem with 4:2:0 (which we are currently using
for most content) is that it is still too heavy on the
CPU load to make streaming of many videos in parallel
completely smooth, which is our primary requirement.

Simply using uncompressed actually did not occur to me
beforehand. Next time we get together I will see how
far we can get with this until the drive's bandwidth
is the bottleneck, which I fear might become an issue
quickly. Thanks for having a look at this!

Best regards,
-- 
Patric Schmitz <bzk0711 at aol.com>