Frédéric Branchaud-Charron
Frédéric Branchaud-Charron
Sorry I wasn't clear enough. The original instance doesn't change, but the instance per worker will. As I said, probably not the right behavior.
This is a bug in Pillow, they are already working on a fix. https://github.com/python-pillow/Pillow/pull/5572
Could you add a test to validate this behaviour?
I think this will be solved in the new design? See https://github.com/keras-team/governance/pull/6/files I know that this proposition is not yet accepted, but once it is, it should be pretty fast.
We use scipy here for SVD (I think np.linalg.svd is called?) [here](https://github.com/keras-team/keras-preprocessing/blob/54b5d6ed65f1608ca7207d585075c2ad3a52929e/keras_preprocessing/image/image_data_generator.py#L986) And we use scipy.ndimage for matrix multiplication here: [ndimage.interpolation.affine_transform](https://github.com/keras-team/keras-preprocessing/blob/54b5d6ed65f1608ca7207d585075c2ad3a52929e/keras_preprocessing/image/affine_transformations.py#L387). Do you think it would require a lot of...
PRs welcome then :)
Hello, We only support 32-bits integers images, in your cases, the images are floats I think. If you need more flexibility, maybe tf.data will help you?
Could you please try with keras master branch?
Hi, We could work on this better if you could provide a standalone reproducible example. Some advices: * Don't use custom data or custom paths. * Use random arrays or...
* This would be a major breaking change for tons of applications. * This would be a pain for Windows users since they need Visual C++. * This is also...