#image-artistic-style-similarity

After # compute styles when calling "style" facing an issue:

InvalidArgumentError: Must provide as many biases as the last dimension of the input tensor: [3] vs. [1,400,400,4] [Op:BiasAdd] name: style_model_7/BiasAdd/.

I found a reason initially if I run this:

compute styles

image_style_embeddings = {}
for image_path in tqdm(image_paths):
image_tensor = load_image(image_path)
style = style_to_vec(image_to_style(image_tensor))
image_style_embeddings[ntpath.basename(image_path)] = style

Then for some reason, the shape remains [1,400,400,4] when the tensor is empty but after when tensor start to fill the shape changes to [1,400,400,3]

But if I run this:

compute styles

image_style_embeddings = {}
for image_path in tqdm(image_paths):
->image_tensor = load_image(image_path)
->if (image_tensor.shape[3]!= 4):
-> ->style = style_to_vec(image_to_style(image_tensor))
->->image_style_embeddings[ntpath.basename(image_path)] = style

Its working fine is it ok to do?

Thanks in advance!