🐞fix support for non-square images #204
Conversation
| @@ -40,13 +42,14 @@ def __init__( | |||
| ): | |||
| super().__init__() | |||
|
|
|||
| assert input_size % 16 == 0, "Input size should be a multiple of 16" | |||
| assert input_size[0] % 16 == 0 and input_size[1] % 16 == 0, "Input size should be a multiple of 16" | |||
There was a problem hiding this comment.
I guess this is no longer needed now that we use padding to make sure that the network can process any image size.
| """ | ||
| # find the largest dimension | ||
| l_dim = 2 ** math.ceil(math.log(max(*input_size), 2)) | ||
| padding = math.ceil((l_dim - input_size[0]) // 2 + 1), math.ceil((l_dim - input_size[1]) // 2 + 1) |
There was a problem hiding this comment.
This won't work for odd image size, e.g. 221
| return torch.mean(torch.pow((latent_i - latent_o), 2), dim=1).view(-1) # convert nx1x1 to n | ||
|
|
||
| def get_padded_tensor(self, batch: Tensor) -> Tensor: |
There was a problem hiding this comment.
Minor issue, but I don't like this name. Maybe apply_padding or pad_inputs? I also think the method could be static.
There was a problem hiding this comment.
I'll update it in a bit. Static is a good idea. In that case would it make sense to move it to data/utils/image.py?
There was a problem hiding this comment.
Yes, I think it would make sense to move it to a shared location so that other modules could access it as well. In that case we should maybe name it pad_nextpow2 or similar
anomalib/data/utils/image.py
Outdated
| """Compute required padding from input size and return padded images. | ||
|
|
||
| Finds the largest dimension and computes a square image to pass into the model. In case the image dimension |
There was a problem hiding this comment.
Maybe mention in the description that the function pads the images so that both sides are a power of 2.
Description
Computes padding based on the image size. Might not be the best approach.
Fixes GANomaly breaks when changing image size #136
Changes
Checklist