Random_StyleIAMDataset bug?

[vincentkk6356]

One-DM/data_loader/loader.py

Lines 203 to 217 in bc705c0

	def get_style_ref(self, wr_id): # Choose the style image whose length exceeds 32 pixels
	style_list = os.listdir(os.path.join(self.style_path, wr_id))
	random.shuffle(style_list)
	for index in range(len(style_list)):
	style_ref = style_list[index]
	style_image = cv2.imread(os.path.join(self.style_path, wr_id, style_ref), flags=0)
	laplace_image = cv2.imread(os.path.join(self.laplace_path, wr_id, style_ref), flags=0)
	if style_image.shape[1] > 128:
	break
	else:
	continue
	style_image = style_image/255.0
	laplace_image = laplace_image/255.0
	return style_image, laplace_image

Why is this function looping all the images instead of just picking one? It slows down the inference by quite a lot.

[dailenson]

This function isn't looping through all the images. It stops once it finds an image with a width greater than 128 pixels. The reasoning behind this is that longer images help ensure more stable performance, as mentioned in Section I of our paper.