Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

color space fix for ImageIO backend (OS X, iOS) #86

Closed
SDLBugzilla opened this issue Feb 11, 2021 · 0 comments
Closed

color space fix for ImageIO backend (OS X, iOS) #86

SDLBugzilla opened this issue Feb 11, 2021 · 0 comments

Comments

@SDLBugzilla
Copy link
Collaborator

@SDLBugzilla SDLBugzilla commented Feb 11, 2021

This bug report was migrated from our old Bugzilla tracker.

These attachments are available in the static archive:

Reported in version: 2.0.0
Reported for operating system, platform: Mac OS X 10.8, x86_64

Comments on the original bug report:

On 2014-11-20 00:10:37 +0000, Jeffrey Carpenter wrote:

Created attachment 1935
proposed colorspace fix for ImageIO backend

I believe I may have found the origin of a serious bug in the ImageIO implementation of SDL_image, where it inaccurately returns pixel color values that are off by one -- i.e.: R(110), G(143), B(189) instead of R(110), G(144), B(190). I've been using a custom built SDL_image with libpng & friends for the backend to work around this issue.

Cheers,
Jeffrey Carpenter i8degrees@gmail.com

On 2014-11-20 00:11:53 +0000, Jeffrey Carpenter wrote:

Created attachment 1936
reference test case data (diag_main.txt before patch)

On 2014-11-20 00:12:23 +0000, Jeffrey Carpenter wrote:

Created attachment 1937
comparison test case data (diag_main.txt after patch)

On 2014-11-20 00:12:45 +0000, Jeffrey Carpenter wrote:

Created attachment 1938
RGB8888 (24-bit, BMP) test case image

On 2014-11-20 00:13:01 +0000, Jeffrey Carpenter wrote:

Created attachment 1939
ARGB8888 (32-bit, PNG) test case image

On 2014-11-20 00:14:09 +0000, Jeffrey Carpenter wrote:

Created attachment 1940
test case (for use with test.bmp, test.png to produce diag_main.txt output)

On 2014-11-20 00:18:52 +0000, Jeffrey Carpenter wrote:

Sorry .. I probably should have zipped all that before hand -_- Anyhow ...

Test case resources (should be able to be replaced with your own as long as the pixel format and file encoding matches):
test.bmp RGB8888 (24-bit, BMP)
test.png ARGB8888 (32-bit, PNG)

This patch reverts revision 292 back to the original author's implementation (Eric Wing). The comments made in this revision seem to contradict what is intended; see CGColorSpaceCreateCalibratedRGB [1] and CGColorSpaceCreateDeviceRGB [2] for Apple's API notes. Also note that the revision comments by "Albert" use a test case that is for use with SDL1 -- how this may change things is unknown to me.

  1. https://developer.apple.com/library/ios/documentation/GraphicsImaging/Reference/CGColorSpace/index.html#//apple_ref/c/func/CGColorSpaceCreateCalibratedRGB
  2. https://developer.apple.com/library/ios/documentation/GraphicsImaging/Reference/CGColorSpace/index.html#//apple_ref/c/func/CGColorSpaceCreateDeviceRGB

Additional notes

  • Tested on Mac OS X 10.10 "Yosemite".
  • Has not been tested against iOS, SDL1 or images that have been processed with pre-multiplied alpha.

Build test case:
clang++ sdl2_image-colorspace-fix.cpp -o sdl2_image-colorspace-fix -framework SDL2 -framework SDL2_image

Run test case to generate my existing data (my baseline / reference):
(output in diag_main.txt, using test.bmp and test.png)

bitmap(0, 0) = ff6e8fbd
png(0, 0) = ff6e8fbd
sdlbmp(0, 0) = 6e90be

bitmap(31, 0) = ff6e8fbd
png(31, 0) = ff6e8fbd
sdlbmp(31, 0) = 6e90be

NOTE: ff6e8fbd = R(110) G(143) B(189)
The actual pixel value is 110, 144, 190. SDL_LoadBMP gets it right in both
cases, whereas IMG_Load (ImageIO backend) gets it wrong on both counts.

Apply revert-rev-292.patch:
hg import revert-rev-292.patch

Rebuild test case, ensuring that we compile and run with the proper runtime
search paths for our patched SDL2_image library (I built mine using Xcode project files):

# man dyld (1)
DYLD_FRAMEWORK_PATH=<my_local_prefix_lib_path> ./sdl2_image-colorspace-fix

Comparison output of reference data with patched library:
(output in diag_main.txt, using test.bmp and test.png)

bitmap(0, 0) = ff6e90be
png(0, 0) = ff6e90be
sdlbmp(0, 0) = 6e90be

bitmap(31, 0) = ff6e90be
png(31, 0) = ff6e90be
sdlbmp(31, 0) = 6e90be

NOTE: ff6e90be = R(110) G(144) B(190)
The actual pixel values match using both SDL_LoadBMP and IMG_Load. This is the
expected behavior.

References

  1. Introduction of ImageIO backend
  2. Pixel bug in Mac OS X
  3. Perhaps again? Pixel bug in Mac OS X

On 2014-11-29 20:09:25 +0000, Sam Lantinga wrote:

Fixed, thanks!
https://hg.libsdl.org/SDL_image/rev/55801d955e3a

I'm not sure what the problem was with the original device RGB colorspace. I tested this with the default calibration and a tweaked monitor calibration and albert's original test case worked fine here.

Thanks!

On 2014-11-30 00:20:50 +0000, Jeffrey Carpenter wrote:

Yay! You the man! :D

Your comments make me curious ... I'll see if I can get different results by playing around with my display's calibration -- I've got three of them to test on. I'll report back if I find anything.

When was it that you tested? I'm guessing during round about time of Albert's patch?

On 2014-11-30 06:29:29 +0000, Sam Lantinga wrote:

Albert's original sample test is still live, if you want to grab it. I played with calibration and it didn't affect the test here.

On 2014-12-06 12:35:11 +0000, Jeffrey Carpenter wrote:

Using Albert's sdlimagetest.cpp [1] and the official distributed binaries of SDL_image (AKA Albert's patch), I get the following output:

$ DYLD_FRAMEWORK_PATH=/Users/jeff/Projects/scratch/sdl/sdl2/sdl2_image-colorspace-fix/libs/before ./sdlimagetest
PixelFormat:
BitsPerPixel: 32, BytesPerPixel: 4
R/G/B/A mask: ff0000/ff00/ff/ff000000
R/G/B/A loss: 0/0/0/0
Colorkey: 0, Alpha: 255
FF0500FF
FF0500FF
FF0500FF
FF0500FF
FF0500FF
FF0500FF
FF0500FF
FF0500FF
FF0500FF
FF0500FF

FF0500FF is RGBA(255, 5, 0, 255). Note that this differs from the expected "bad" output of SDL_image, which expects FF1514F4.

Using Albert's sdlimagetest.cpp [1] with my proposed patch built in SDL_image, I get the following output:

$ DYLD_FRAMEWORK_PATH=/Users/jeff/Projects/scratch/sdl/sdl2/sdl2_image-colorspace-fix/libs/after ./sdlimagetest
PixelFormat:
BitsPerPixel: 32, BytesPerPixel: 4
R/G/B/A mask: ff0000/ff00/ff/ff000000
R/G/B/A loss: 0/0/0/0
Colorkey: 0, Alpha: 255
FF0000FF
FF0000FF
FF0000FF
FF0000FF
FF0000FF
FF0000FF
FF0000FF
FF0000FF
FF0000FF
FF0000FF

FF0000FF is RGBA(255, 0, 0, 255). Note that this is what Albert's test expected, too, with his patch.

Choosing different color calibrated color profiles at random made no difference whatsoever for me in any of these tests.

I think Albert's test case was built with SDL1 in mind, whereas my tests are with SDL2. My tests also use a newer version of SDL_image than at the revision where Albert's changes were introduced. I don't know how this might change any of the results.

In summary, I don't know either! The conflicting test results leave me ever more... confused, with more questions than answers. Such is often the case :-)

In any case, the fact that my proposed patch doesn't appear to introduce problems for either one of us is hopefully a good sign of things. I haven't yet found anybody else to help test this with, but it's only a matter of time! In conclusion, until more information comes to light, I feel OK about my proposed patch.

Thanks,
Jeffrey Carpenter i8degrees@gmail.com

  1. https://hg.libsdl.org/SDL_image/rev/5953114c0d27
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant