You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Aug 5, 2022. It is now read-only.
I also tested this issue with scaled input video to match 16x16 or 64x64 raster (720x480 and 768x512) with the same result.
my explanation is something like this - intel-vaapi-driver is doing 1 pass encoding and the estimator for cbr/vbr bitrate usage runs out of useable bandwidth capacity.
This might be related to the fact that libyami VP8 SVCT implementation is using only last and golden references and not alternate frame references. This might influence the cbr/vbr calculation, but I can't confirm this information.
The text was updated successfully, but these errors were encountered:
ottingerg
changed the title
VP8 SVVT yamitranscode pink artefakts with fast moving picture in CBR/VBR mode
VP8 SVCT yamitranscode pink artefakts with fast moving picture in CBR/VBR mode
Jun 1, 2018
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.
I do get pink artefacts when using vp8 temporal scaling (SVCT) with fast moving pictures in CBR and VBR mode.
Example "football (a)" from:
https://media.xiph.org/video/derf/
Command Line:
./yamitranscode -i ~/Videos/football_w720_h486_f360_fps30.yuv -W 720 -H 486 -o football_cbr700_svct3_upstream.ivf -c VP8 --btl0 700 --btl1 700 --intraperiod 32 --rcmode CBR -b 700 -t 3
I also tested this issue with scaled input video to match 16x16 or 64x64 raster (720x480 and 768x512) with the same result.
my explanation is something like this - intel-vaapi-driver is doing 1 pass encoding and the estimator for cbr/vbr bitrate usage runs out of useable bandwidth capacity.
This might be related to the fact that libyami VP8 SVCT implementation is using only last and golden references and not alternate frame references. This might influence the cbr/vbr calculation, but I can't confirm this information.
The text was updated successfully, but these errors were encountered: