adobe/antialiased-cnns
A PyTorch library providing antialiased CNN models (resnet, vgg, densenet variants) with BlurPool layers for improved shift-invariance in image classification.
Not currently ranked — collecting fresh signals.
star history
This repository implements the antialiased CNN technique from the ICML 2019 paper ‘Making Convolutional Networks Shift-Invariant Again’. It provides a BlurPool layer that replaces standard downsampling operations in CNNs to reduce aliasing artifacts. The library includes pretrained antialiased versions of common architectures (ResNet, VGG, DenseNet, ResNeXt, Wide ResNet) that achieve better stability and accuracy on image classification tasks. Users can either load pretrained models directly or integrate the BlurPool layer into their own custom architectures.
Frequently asked
- What is adobe/antialiased-cnns?
- A PyTorch library providing antialiased CNN models (resnet, vgg, densenet variants) with BlurPool layers for improved shift-invariance in image classification.
- Is antialiased-cnns open source?
- Yes — adobe/antialiased-cnns is an open-source project tracked on heatdrop.
- What language is antialiased-cnns written in?
- adobe/antialiased-cnns is primarily written in Python.
- How popular is antialiased-cnns?
- adobe/antialiased-cnns has 1.7k stars on GitHub.
- Where can I find antialiased-cnns?
- adobe/antialiased-cnns is on GitHub at https://github.com/adobe/antialiased-cnns.