This repository has been archived by the owner on Nov 22, 2022. It is now read-only.
implemented bottleneck separable convolutions #855
Closed
Conversation
This file contains 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
facebook-github-bot
added
the
CLA Signed
shreydesai
force-pushed
the
export-D16563566
branch
from
db40d89 to
4409772
Compare
shreydesai
added a commit
to shreydesai/pytext
that referenced
this pull request
Aug 2, 2019
Summary: Pull Request resolved: facebookresearch#855 Creates bottleneck layers for separable convolutions. Downsampling, convolving, and then upsampling significantly cuts the number of parameters with minimal loss in performance. This diff is a variant of the traditional bottleneck method -- instead of upsampling directly in the pointwise convolution, we split it into two pieces where the first convolution downsamples into a (sufficiently small) low dimension and the second convolution upsamples into the target (higher) dimension. Example: Given an input with 256 channels, the depthwise channel spatially convolves it with 256 channels. Then, instead of projecting the channel space to 256 with the pointwise convolution, it is split into two pieces. The first pointwise convolution projects the channel space to 64, then the second pointwise convolution projects the channel space to 256 -- as we intended. {F172681616} Differential Revision: D16563566 fbshipit-source-id: e742abce052380a5f1174a4180276fdbbacc5b41
shreydesai
force-pushed
the
export-D16563566
branch
from
4409772 to
e34a891
Compare
shreydesai
added a commit
to shreydesai/pytext
that referenced
this pull request
Aug 2, 2019
Summary: Pull Request resolved: facebookresearch#855 Creates bottleneck layers for separable convolutions. Downsampling, convolving, and then upsampling significantly cuts the number of parameters with minimal loss in performance. This diff is a variant of the traditional bottleneck method -- instead of upsampling directly in the pointwise convolution, we split it into two pieces where the first convolution downsamples into a (sufficiently small) low dimension and the second convolution upsamples into the target (higher) dimension. Example: Given an input with 256 channels, the depthwise channel spatially convolves it with 256 channels. Then, instead of projecting the channel space to 256 with the pointwise convolution, it is split into two pieces. The first pointwise convolution projects the channel space to 64, then the second pointwise convolution projects the channel space to 256 -- as we intended. {F172681616} Differential Revision: D16563566 fbshipit-source-id: bff549728803062045a8540b069791e96c7944f1
shreydesai
force-pushed
the
export-D16563566
branch
from
e34a891 to
3f5074c
Compare
shreydesai
added a commit
to shreydesai/pytext
that referenced
this pull request
Aug 2, 2019
Summary: Pull Request resolved: facebookresearch#855 Creates bottleneck layers for separable convolutions. Downsampling, convolving, and then upsampling significantly cuts the number of parameters with minimal loss in performance. This diff is a variant of the traditional bottleneck method -- instead of upsampling directly in the pointwise convolution, we split it into two pieces where the first convolution downsamples into a (sufficiently small) low dimension and the second convolution upsamples into the target (higher) dimension. Example: Given an input with 256 channels, the depthwise channel spatially convolves it with 256 channels. Then, instead of projecting the channel space to 256 with the pointwise convolution, it is split into two pieces. The first pointwise convolution projects the channel space to 64, then the second pointwise convolution projects the channel space to 256 -- as we intended. {F172681616} Reviewed By: geof90 Differential Revision: D16563566 fbshipit-source-id: dd44d8eebea00dbb33130febd6fc7bda9d735aa7
Summary: Pull Request resolved: facebookresearch#855 Creates bottleneck layers for separable convolutions. Downsampling, convolving, and then upsampling significantly cuts the number of parameters with minimal loss in performance. This diff is a variant of the traditional bottleneck method -- instead of upsampling directly in the pointwise convolution, we split it into two pieces where the first convolution downsamples into a (sufficiently small) low dimension and the second convolution upsamples into the target (higher) dimension. Example: Given an input with 256 channels, the depthwise channel spatially convolves it with 256 channels. Then, instead of projecting the channel space to 256 with the pointwise convolution, it is split into two pieces. The first pointwise convolution projects the channel space to 64, then the second pointwise convolution projects the channel space to 256 -- as we intended. {F172681616} Reviewed By: geof90 Differential Revision: D16563566 fbshipit-source-id: 1cf310315739802aa5ea6a34efb4c8ab771a3c63
shreydesai
force-pushed
the
export-D16563566
branch
from
3f5074c to
d39c777
Compare
|
This pull request has been merged in 352b8be. |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
Summary: Creates bottleneck layers for separable convolutions. Downsampling, convolving, and then upsampling significantly cuts the number of parameters with minimal loss in performance. This diff is a variant of the traditional bottleneck method -- instead of upsampling directly in the pointwise convolution, we split it into two pieces where the first convolution downsamples into a (sufficiently small) low dimension and the second convolution upsamples into the target (higher) dimension.
Differential Revision: D16563566