Michael Gharbi

Tzu-Mao Li might know better, but we have not tested the scheduler on OpenCL GPU hardware, only CUDA at the moment. There might be a mismatch between the size of...

Hi, The difference inside / outside photoshop is most likely due to an unspecified ICC profile (i.e. your image is linear, but the skimage.io output marks it as encoded in...

Hi, Yes the cv2 write should write the correct linear data to file (though it will most likely assign it an sRGB profile, which may make it look strange in...

Hum, this is odd, but your model does not seem to have converged yet. On the positive side, the validation and training curves are on-par so you’re not overfitting. Do...

One thing I can think of is the guidance map could be degenerated, which would lead to some unused bins in the bilateral grid. You can debug this by visualizing...

I unfortunately could not reproduce this error on gcc-5.0. Does adding the -D_GLIBCXX_USE_CXX11_ABI=0 flag help? I'll add it to the Makefile.

Merci Thomas, Feel free to pull-request your updates, otherwise I'll add those changes to the Makefile and close the issue.

Dear vikasrs, I could not retrieve the original StyleTransferN model from a codebase I can publicly release, sorry about that. -- Michaël From: vikasrs Reply-To: mgharbi/hdrnet_legacy Date: Monday, October 8,...

That should not matter, those layers have no batchnorm. is_training switches between using the running mean/std (at train time) vs. using the memorized values (at test time).

The expert models reproduce the retouches of human photographer as described in the paper. Variations like _cm1 refer to “channel multiplier” which were used to explore performance trade-offs. I included...