Richard Zhang

An example for extracting the global color histogram is [here](https://github.com/junyanz/interactive-deep-colorization/blob/master/DemoGlobalHistogramTransfer.ipynb) in function `get_global_histogram`. It called a network [here](https://github.com/junyanz/interactive-deep-colorization/blob/master/models/global_model/global_stats.prototxt). The network resizes the image, computes the color bin that each pixel...

I believe I did not do the initialization properly in the pytorch code. I am looking into it. Thanks for your patience.

The PyTorch version is an approximate reimplementation and performs a bit differently. I will have some more time after CVPR deadline to look into it further. Thanks for your patience.

My apologies for the delay. It is on the list of todo's but I have been caught up with many things. I will get to it as soon as possible.

I'm working on a pytorch version for public release. Thanks for your patience.

Sorry for the delay. I am retraining it in pytorch now. Hopefully release in a few weeks. Thanks for your patience.

A PyTorch reimplementation is in this [repository](https://github.com/richzhang/colorization-pytorch/tree/master). We updated this repository as well to support PyTorch models in the backend. Apologies again for the delay.

We are working on releasing training code. Thanks for your patience!

Sorry for the big delay. We reimplemented pytorch training code [here](https://github.com/richzhang/colorization-pytorch).

@off99555 Yes, `Conv (stride2)+Relu` should be replaced with `Conv(stride1)+Relu+BlurPool(stride2)`. See the readme or [antialiased resnet](https://github.com/adobe/antialiased-cnns/blob/master/models_lpf/resnet.py) for reference.