RandomCrop bugs

[martin-gorner]

Repro Colab here

Input image size: 360*512px

Problem #1: cropping to 128x513 crashes with a div by 0
Problem #2: cropping to 128x514 no longer produces random results (there are still many possible random positions along the Y axis)

[bhack]

Yes for # 1 it is an edge case of get_random_transformation in Keras
https://github.com/keras-team/keras/blob/v2.10.0/keras/layers/preprocessing/image_preprocessing.py#L581-L582
w_start = rands[1] % (w_diff + 1)

As your image in the colab cell is

tf.Tensor([372 512   3], shape=(3,), dtype=int32)

With

random_crop = tf.keras.layers.RandomCrop(height=128, width=513)

You are going to have a w_diff of -1 and so you crash with % 0

[bhack]

For # 2 I don't understand well the logic of using dtype.max
https://github.com/keras-team/keras/blob/v2.10.0/keras/layers/preprocessing/image_preprocessing.py#L580

rands = self._random_generator.random_uniform([2], 0, dtype.max, dtype)

[LukeWood]

Thank you Martin for the bug report!

[sachinprasadhs]

Thanks for reporting the issue! We have consolidated the development of KerasCV into the new KerasHub package, which supports image, text, and multi-modal models. Please read the announcement. KerasHub will support all the core functionality of KerasCV.

KerasHub can be installed with !pip install -U keras-hub. Documentation and guides are available at keras.io/keras_hub.

With our focus shifted to KerasHub, we are not planning any further development or releases in KerasCV. If you encounter a KerasCV feature that is missing from KerasHub, or would like to propose an addition to the library, please file an issue with KerasHub.