Use 1x1 convolutions to improve BinaryAlexNet latency on Compute Engine #320
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
lgeiger
commented
May 26, 2021
| @@ -91,9 +92,9 @@ def build(self) -> tf.keras.models.Model: | |||
| if self.include_top: | |||
| weights_path = utils.download_pretrained_model( | |||
| model="binary_alexnet", | |||
| version="v0.2.0", | |||
| version="v0.3.0", | |||
| file="binary_alexnet_weights.h5", | |||
There was a problem hiding this comment.
These weights have been manually reshaped from previous model so that Keras can correctly reload them.
jneeven
approved these changes
May 26, 2021
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in 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Outdated suggestions cannot be applied.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.
This PR updates
BinaryAlexNetto use 1x1 convolutions in the classification layer which allows correct conversion to LCE.This brings the official benchmarks inline with the numbers we quote in the LCE paper. This has a big impact since the classification layers are actually very large.
This code has been lying around on my system for a long time now, so I briefly spent a few minutes converting the weights which was just a matter of reshaping the weights of the last four classification layers.