Error when training about adain-tf HOT 4 CLOSED

This is an issue with the torchfile package on Windows. Someone provided a fix for my WCT repo that I should port over to this one -- you can access the fixed compatible version of torchfile.py here https://github.com/eridgd/WCT-TF/blob/master/torchfile.py to use instead of the pip version.

from adain-tf.

Thanks heaps. Also I had to copy the vgg_normalized.py from your WCT-TF as it had some edits as well?
Cheers

from adain-tf.

I'm trying to train using a template, I keep encountering this error:

ufunc 'true_divide' output (typecode 'd') could not be coerced to provided output parameter (typecode 'l') according to the casting rule ''same_kind''

I have narrowed it down to the torchfile.py that I downloaded from your link above in this comment thread. But beyond that, I'm not able to tell if the error is in the read_double() or the read_long() functions or how I can explicitly say to not type-check when casting, as I don't see a place where explicit casting is happening.

from adain-tf.

From a quick pass of looking into this, my best guess that this is due to the version of numpy being used. Any luck when you try downgrading np?

Also, are you trying this on Windows? If not, the official version of torchfile (or at least the version from 11 months ago) should work on linux/macos.

from adain-tf.