Gabriel Rasskin

Rescaling layer defaults to "float32" if dtype not specified - input is uint8

2

Backwards incompatibility from [semisupervised_simclr.py](https://github.com/keras-team/keras-io/blob/master/examples/vision/semisupervised_simclr.py) Input image is `tf.Tensor(shape=(8, 96, 96, 3), dtype=uint8)` and is unable to multiply by float tensor in rescaling layer. Current workaround is to add dtype via...

Layer does not have a default tensorflow namespace; requires passing `initial_value` in lambda function

1

Ran into this with swim_transformer.py example. We create a custom layer that uses numpy ops and a tensorflow variable to store state. ``` self.relative_position_index = tf.Variable( initial_value=tf.convert_to_tensor(relative_position_index), trainable=False ) ```...