What steps will reproduce the problem?
1. Take a file (tiff for example
2. compress it with cwebp using the -lossless parameter and different values
for -q and -m
3. the resulting file size differs with little chance to pre-select optimal
values
What is the expected output? What do you see instead?
https://docs.google.com/spreadsheet/oimg?key=0AvPoCfoFzhyUdE9kd1V5MkMzLUpMZHZBSU
c0RmR0bHc&oid=3&zx=h2t8wowl9hje is a plot for a single file that has been
lossless compressed 707 times (101 times (q=0 to q=100 and m=0 to m=6). q=0
will produce the largest resulting file sizes and m=6 will generally be smaller
than m=5. The file size for m=1, m=2 and m=3 was identical for all q values in
my test case. (number of files tested: 1)
Smalles files were
3510686 test-lossless-m5-q72.webp
3510686 test-lossless-m5-q73.webp
3510686 test-lossless-m6-q72.webp
3510686 test-lossless-m6-q73.webp
3510796 test-lossless-m5-q98.webp
3510796 test-lossless-m5-q99.webp
3510796 test-lossless-m6-q98.webp
3510796 test-lossless-m6-q99.webp
...
Largest files were
...
3558452 test-lossless-m0-q85.webp
3558512 test-lossless-m0-q34.webp
3558512 test-lossless-m0-q35.webp
3558524 test-lossless-m0-q55.webp
3558540 test-lossless-m0-q68.webp
3558576 test-lossless-m0-q64.webp
3558576 test-lossless-m0-q65.webp
3558642 test-lossless-m0-q71.webp
3558788 test-lossless-m0-q36.webp
3565922 test-lossless-m5-q00.webp
3565922 test-lossless-m6-q00.webp
3580364 test-lossless-m4-q00.webp
3597430 test-lossless-m1-q00.webp
3597430 test-lossless-m2-q00.webp
3597430 test-lossless-m3-q00.webp
3612804 test-lossless-m0-q00.webp
The expected behaviour of cwebp would be to provide the smallest possible
lossless file size with a predictable set of parameters.
What version of the product are you using? On what operating system?
webp 0.3.0 on x86/cygwin/Windows 8 compliled with gcc
Please provide any additional information below.
The original file was a 1794*1743 TIFF image with a size of 7149230 bytes,
developed from a Nikon NEF raw file. The file is licensed under a free creative
commons license and can be provided on request to allow reproduction.
Original issue reported on code.google.com by mathias.schindler on 20 May 2013 at 10:27
Original issue reported on code.google.com by
mathias.schindleron 20 May 2013 at 10:27