tests: Add visual/define_bits_jpeg2 test #21738
Conversation
kjarosh
added
A-tests
|
Now this is thorough, as always - hats off to you! 🎩 |
|
One thing to consider - with JPEG having 8x8 blocks for IDCT, the two bottom rows and two rightmost columns will be on different blocks. On some images, you can even see some discoloration on the next-to-last column because of this. |
|
Also, did you make the expected output by screenshotting Flash Player, or by stitching together the input files in an image editor? |
| @@ -0,0 +1,18 @@ | |||
| num_ticks = 1 | |||
|
|
|||
| # FIXME We're badly decoding JPEG images with 4:2:0 and 4:4:0 chroma subsampling | |||
There was a problem hiding this comment.
If the test pattern was rotated by 90 degrees, perhaps some other subsampling options would (also) look wrong!
There was a problem hiding this comment.
Always those chroma sample locations...
https://www.pcmag.com/encyclopedia/term/chroma-subsampling
https://www.sciencedirect.com/topics/engineering/chroma-sample
Notice how there is a difference of horizontal and vertical chroma sample offsets in the different modes.
What I'm also saying is - if the test pattern was magnified just 2x (with nearest interpolation, or even recreated with two-pixel tall gradients) - it would probably look a lot less wrong, even if there's technically a mismatch. And not a lot of real world content is like these images.
This covers image decoding code paths in Ruffle.
kjarosh
force-pushed
the
define-bits-test
branch
from
fdbf8c3 to
6839108
Compare
|
Added a readme, I'll look into including other patterns too |
My outputs are always from Flash Player. In this case I made sure it's the same on different platforms too. |
|
Neat! But guess what - this is what my FP outputs: 
Difference from the expected output being: 
|
2 participants
This covers image decoding code paths in Ruffle.