Changing to AdaptiveAvgPool2d on SqueezeNet and ResNet #643
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The Average Pool Layer on SqueezeNet and ResNet was hardcoded, this was changed, now any the input size is accepted.
|
@houseroad will this change break ONNX? I'm inclined to make all torchvision models take an |
houseroad
reviewed
Oct 31, 2018
There was a problem hiding this comment.
https://github.com/pytorch/pytorch/blob/master/torch/onnx/symbolic.py#L599
(1,1) case is supported. I have manually tested SqueezeNet, it works fine. I think this change will NOT break ONNX test.
Thanks.
This was referenced Nov 6, 2018
|
Thanks to you and your work!! |
This was referenced Nov 27, 2018
This was referenced Feb 13, 2019
fmassa
pushed a commit
that referenced
this pull request
Feb 14, 2019
The update allows Alexnet to process images larger or smaller than prescribed imagenet size using adaptive average pooling. Will be useful while finetuning or testing on different resolution images. Similar to #643 and #672. I did not include adaptive avg pool in features or classifier block so that these predefined blocks can be used as it is.
fmassa
pushed a commit
that referenced
this pull request
Feb 14, 2019
The update allows VGG to process images larger or smaller than prescribed imagenet size using adaptive average pooling. Will be useful while finetuning or testing on different resolution images. Similar to #643 and #672. I did not include adaptive avg pool in features or classifier block so that these predefined blocks can be used as it is.
# for free
to join this conversation on GitHub.
Already have an account?
# to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Hi,
I'm dealing with this problem with the hardcoded AvgPool2d on SqueezeNet, some people notice this a year ago, also in ResNet, so I made these changes, Pytorch must be as robust as possible. I left the other AvgPool2d layers because they have another purpose.
nn.AvgPool2d -> nn.AdaptiveAvgPool2d((1,1)) assures the output's size like (batch_size, num_classes, 1,1) on SqueezeNet and ResNet.