Evan Davis

It should be close. I tried to stay faithful to the architecture & training process as described in the paper. I would also recommend taking a look at this TensorFlow...

It's been too long for me to remember exactly why I implemented things the way I did, but my guess is it's a mistake. The results do still seem to...

Hi, The normalized VGG differs in three ways from standard VGG19: * Pre-processing is done as part of the network in the first layer. It expects RGB in [0,1], scales...

Unfortunately not and don't have plans to do so soon, but two repos I've seen that do implement Spatial Control are: * https://github.com/ftokarev/tf-adain * https://github.com/Yijunmaverick/UniversalStyleTransfer#spatial-control

I don't seem to have a trained model any more, I'll re-train when I can spare the GPU and post it here.

@Alan-xw could you please post a link to your code?

The original VGG19 was trained on images loaded as BGR from OpenCV and so expects that channel order. See https://github.com/jcjohnson/neural-style/issues/207 The second key preprocessing step of the VGG nets is...

You need to have an additional dimension for batch, even with just one image: ``` img = load_img(args.image) assert img.ndim == 3 img = np.expand_dims(img, axis=0) # Alternatively could do:...