Closed bat999 closed 3 years ago
No, I'm deliberately hiding the quantizer option.
As far as I can tell min/max split didn't do anything for still images.
The scale is not portable across AV1 encoders and the number doesn't mean anything that non-experts can relate to.
The effect on visual quality is non-linear. My current implementation is wrong actually. I plan to align it better to approximate JPEG quality scale.
As far as I can tell min/max split didn't do anything for still images. Yes, I can understand that. Thanks.
... I plan to align it better to approximate JPEG quality scale. Both cjpeg and cwebp have default quality 75. dssim thinks cavif Q 75 is too good. :-)
0.00164695 foo1_q75.jpg
0.00244810 foo1_q75.webp
0.00084624 foo1_q75.avif
0.00203879 foo2_q75.jpg
0.00256523 foo2_q75.webp
0.00082852 foo2_q75.avif
echo ""; \
cjpeg -quality 75 -outfile foo1_q75.jpg foo1.png; \
cwebp -q 75 -mt -m 6 -quiet foo1.png -o foo1_q75.webp; \
cavif -Q 75 -s 0 -j $(nproc) -q -o foo1_q75.avif foo1.png; \
dssim foo1.png foo1_q75.jpg foo1_q75.webp foo1_q75.avif; \
echo ""; \
cjpeg -quality 75 -outfile foo2_q75.jpg foo2.png; \
cwebp -q 75 -mt -m 6 -quiet foo2.png -o foo2_q75.webp; \
cavif -Q 75 -s 0 -j $(nproc) -q -o foo2_q75.avif foo2.png; \
dssim foo2.png foo2_q75.jpg foo2_q75.webp foo2_q75.avif
Hi I can see you have gone to a lot of trouble to map Q to the rav1e quantizers. This makes it intuitive for people who use cjpeg and cwebp etc. When they have experimented some they will settle on a value of Q to use for cavif.
But might there be a case to allow access to the quantizers INSTEAD of the Q scale?
Something like this...