myint / perceptualdiff Goto Github PK

windows binary can be downloaded here https://sourceforge.net/projects/pdiff/ but running perceptualdiff.exe i got this error.

should i compile it ?

thanks

Hi - just wondering if you could explain how I could use this fork to build a binary that includes FreeImage (just as they do here http://sourceforge.net/projects/pdiff/files/pdiff/perceptualdiff-1.1.1/)

Following your instructions in the readme does not give me a binary that I can distribute - it always requires FreeImage to be installed on the target.

Thanks :)

qa-Latitude-E6510 Scrivania # perceptualdiff --verbose immagine1.png immagine2.png
Field of view is 45 degrees
Threshold pixels is 100 pixels
The Gamma is 2.2
The Display's luminance is 100 candela per meter squared
Converting RGB to XYZ
Constructing Laplacian Pyramids
Performing test
PASS: Images are perceptually indistinguishable
0 pixels are different
qa-Latitude-E6510 Scrivania #

but the two images are different.

Hi

This is a feature requests, but if you have a great idea on doing it I would be very happy.

Is there a way I can fingerprint an image so I can compare fingerprinting instead of comparing whole images all the time, this is for nearly similar images, the identical have been taken care of with sha-hashing.

I'm trying to find near duplicates in a 3mill+ image database using preceptualdiff --threshold 50000 --scale it works fine, but very slow. We had a SAN crash and found backup in archive mode (means adding, not moving when things have been rewritten, giving us 600K obsolete duplicates).

I've done heavy optimizing and is down to 80K root path images that need to be compared for near duplicates (mostly scaled and color and alpha tunes and some images in path removed), but it is still (80K+80K)/2 = 3.2Bill comparisons to be made. The SSD san manages to run 16 threads in parallel giving me 32 comparisons/sec which will take 542 days.

Introduced in #16

This Pull request effectively introduced a regression in perceptual diff, since this was a feature of version 1.1:

1.1 - Added colorfactor and downsample options. Also always output difference file if requested. Always print out differing pixels even if the test passes.

The Debian autobuilders build each package for various architectures. In the case of the perceptualdiff package, I enabled running the test script test/run_tests.bash during the build, in version 2.1-3, which includes changes beyond v2.1 up to 2a849e1 (6-Aug-2020). The tests succeed on some architectures (amd64, i386, arm64, ppc64el) but fail on others (armel, armhf, s390x).

I have no idea why. Likely something simple, like word size or endianity. Help / ideas welcome!

Details: https://buildd.debian.org/status/package.php?p=perceptualdiff

Is there some simple way to normalize sum_error so it fits into a given range, e.g [0, 1]? Assuming 10x10 RGB 8-bit input images, is 10 * 10 * 255 * 3 the upper bound for sum_error?

Any thoughts into making this into a docker image, so it's really easy to test/play with?

I just tried and failed :( I'm soooo not experienced with c++ etc... so I've messed something up, obviously.

So yeah, any thoughts please?

Does it? Or will it be added? Base64 input is fine too.