Skip to content

Conversation

@DonghweeYoon
Copy link

I got the same issue with #155.

In the current code, the SIFID is calculated from the activation before the first pooling layer.
To calculate the SIFID as described in the paper, the following modifications are required.

self.last_needed_block = max(output_blocks)

If 'dims' is set to 64, 'self.last_needed_block' is set to 0.

# Block 0: input to maxpool1
block0 = [
inception.Conv2d_1a_3x3,
inception.Conv2d_2a_3x3,
inception.Conv2d_2b_3x3,
]
self.blocks.append(nn.Sequential(*block0))
# Block 1: maxpool1 to maxpool2
if self.last_needed_block >= 1:
block1 = [
nn.MaxPool2d(kernel_size=3, stride=2),
inception.Conv2d_3b_1x1,
inception.Conv2d_4a_3x3,
]
self.blocks.append(nn.Sequential(*block1))

Then the input feature map only passed through 'block0'.
It doesn't go through the first max-pooling layer.

Sorry for the duplicate PRs. #156 #157

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant