Lukas Geiger

Glorot normal [draws samples from a truncated normal distribution centered on 0](https://www.tensorflow.org/api_docs/python/tf/keras/initializers/GlorotNormal), so in this case it means binary weights will be randomly set to either -1 or 1.

> Are both Conv2d and QuantConv2D all set to either -1 or 1? Only weights of `QuantConv2D` layers will be binarized.

You can also checkout https://github.com/plumerai/larq/issues/95 to follow our reasoning on why we opted for TensorFlow instead of a different framework. In any case we're happy to answer any questions you...

We use [TensorFlow Datasets](https://www.tensorflow.org/datasets) for preparing the ImageNet dataset. If I recall correctly they used to be downloaded automatically. Recently Tensorflow Datasets switched to allow to use already downloaded tar...

> Needs unit tests @Maratyszcza Sorry about that. I wasn't sure whether there is a concrete reason why reshapes are not yet delegated to XNNPACK before adding tests. I updated...

> LG overall, but please follow the structure of depth_to_space_tester (Test method is provided with the datatype). @Maratyszcza Thanks for the pointer, that's much cleaner. I updated the PR.

@Maratyszcza Thanks for the review, I applied the requested changes and updated the PR.

Looks like some Google internal checks failed. Is there anything that I can do to fix this from my side?

This issue has been open for 1.5 years now (including the https://github.com/tensorflow/tensorflow/issues/48149). @fchollet @mattdangerw @k-w-w Is there any chance this will be fixed? I am very happy to look into...

Looking at TensorFlow's bazel config again, it seems like they [specifically enable some warnings on Linux](https://github.com/tensorflow/tensorflow/blob/e6177657b966d4a0d7c9dcff0e87789bfc968da8/.bazelrc#L303-L313). Should we do the same?