PyConvResNet

The PyConvResNet model is based on the “Pyramidal Convolution: Rethinking Convolutional Neural Networks for Visual Recognition” paper.

Architecture overview

This paper explores an alternative approach for convolutional block in a pyramidal fashion.

https://github.com/frgfm/Holocron/releases/download/v0.2.1/pyconv_resnet.png

The key takeaways from the paper are the following:

  • replaces standard convolutions with pyramidal convolutions

  • extends kernel size while increasing group size to balance the number of operations

Model builders

The following model builders can be used to instantiate a PyConvResNet model, with or without pre-trained weights. All the model builders internally rely on the holocron.models.classification.resnet.ResNet base class. Please refer to the source code for more details about this class.

pyconv_resnet50([pretrained, progress])

PyConvResNet-50 from "Pyramidal Convolution: Rethinking Convolutional Neural Networks for Visual Recognition"

pyconvhg_resnet50([pretrained, progress])

PyConvHGResNet-50 from "Pyramidal Convolution: Rethinking Convolutional Neural Networks for Visual Recognition"