The API in scikit image has moved on from using multichannel= as an argument to skimage.exposure.match_histograms. It was deprecated but available in skimage 0.19, removed in 0.20, and we're on 0.21 as of today. On colab this is fine because the default instance you get has a 0.19.3 scikit-image:
The new way is to use the channel_axis argument. It does not break things on colab (because channel_axis is supported and preferred over multichannel in 0.19) and is the only way forward for newer scikit-image version. See patch below. I don't want to fork on gh, make a branch, fix, and file an MR, etc., someone with write-privileges just commit these few lines, thanks! I'll begrudgingly do all this if I have to, let me know.
See here https://github.com/deforum-art/deforum-stable-diffusion/blob/main/helpers/colors.py#L10-L15
The API in scikit image has moved on from using
multichannel=
as an argument toskimage.exposure.match_histograms
. It was deprecated but available in skimage 0.19, removed in 0.20, and we're on 0.21 as of today. On colab this is fine because the default instance you get has a 0.19.3 scikit-image:The new way is to use the
channel_axis
argument. It does not break things on colab (becausechannel_axis
is supported and preferred overmultichannel
in 0.19) and is the only way forward for newer scikit-image version. See patch below. I don't want to fork on gh, make a branch, fix, and file an MR, etc., someone with write-privileges just commit these few lines, thanks! I'll begrudgingly do all this if I have to, let me know.