This repository contains a keras porting of the work "Making Convolutional Networks Shift-Invariant Again" by Richard Zhang, published in ICML 2019, original code in torch available at github.com/adobe/antialiased-cnns. Any work using this code must cite that work. This code is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
The porting of the BlurPool layer comes with an vgg-face pretrained implementation of SENet from github.com/rcmalli/keras-vggface where the strided convolutional layers and the pooling layers use BlurPool. If you wish to add more example architectures or more pre-trained weights, you are welcome!
Run antialiasing.py for testing the BlurPool layer alone. From senet_lpf import SENET for using the senet-lpf model.