Fuse padding layers into (Depthwise) Convolutions

larq / compute-engine

Highly optimized inference engine for Binarized Neural Networks

https://docs.larq.dev/compute-engine

Apache License 2.0

240 stars 33 forks source link

Fuse padding layers into (Depthwise) Convolutions #712

Closed Tombana closed 2 years ago

Tombana commented 2 years ago

What do these changes do?

The MobileNet models in tf.keras.applications have unneeded explicit padding layers. This PR introduces an MLIR pass that fuses such layers into (Depthwise) Convolutions.

How Has This Been Tested?

An MLIR FileCheck test has been added.