Xpra: Ticket #2061: picture encoding heuristics tweaks
Based on log samples from #2029, #2022 and others:
- maybe use x264 ahead of vpx in more cases (changes to "size-efficiency" attribute?): the bandwidth cost of vp8 / vp9 varies more (which hurts bandwidth management) and / or encode faster (currently double that of x264 at similar speed settings)
- take min-speed and min-quality into account when calculation their corollary
- ensure we don't choose YUV444P csc step
- always try to maintain 25fps, drop speed and quality to try to achieve it
- take scaling attribute into account when using the number of pixels/s
- honour fixed quality / speed: heuristics use min values only at the moment
Better unit tests.
Sun, 25 Nov 2018 16:04:08 GMT - Antoine Martin: status, description changed
- status
changed from new to assigned
- description
modified (diff)
Sun, 25 Nov 2018 16:43:07 GMT - Antoine Martin:
Related changesets and tickets:
Sun, 25 Nov 2018 17:35:25 GMT - Antoine Martin: description changed
- description
modified (diff)
Updates:
- r21098: when min-speed is high, lower the quality quicker
- r21104: prefer x264 over vpx
- r21105: raise initial quality but allow fewer soft-expired regions then gradually increase the limit
- r21108: simplify speed target calculation code
- r21110: video pipeline scoring improvements
- r21112: code cleanup, makes debug output much more readable
Tue, 27 Nov 2018 11:53:10 GMT - Antoine Martin:
See also / updates:
Tue, 27 Nov 2018 14:03:58 GMT - stdedos:
Replying to Antoine Martin:
- always try to maintain 25fps, drop speed and quality to try to achieve it
Minor comment: I would say, given "heuristics", I would prefer 15fps with a non-pixelated output (so that output is legible), rather than 25fps with borked quality. If there is no such heuristic in the code.
Tue, 27 Nov 2018 15:05:54 GMT - Antoine Martin:
Minor comment: I would say, given "heuristics", I would prefer 15fps with a non-pixelated output (so that output is legible), rather than 25fps with borked quality. If there is no such heuristic in the code.
There is: just raise the min-quality.
Fri, 11 Jan 2019 13:08:30 GMT - Antoine Martin: status changed; resolution set
- status
changed from assigned to closed
- resolution
set to fixed
Many important heuristics updates recorded in ticket:2090#comment:6 and ticket:2090#comment:7.
Fri, 11 Jan 2019 13:21:50 GMT - Antoine Martin:
And also: ticket:2090#comment:10
Sat, 23 Jan 2021 05:41:08 GMT - migration script:
this ticket has been moved to: https://github.com/Xpra-org/xpra/issues/2061