SRI-Conv 1.2.5

Official package for 'Symmetric Rotation-Invariant Convolutional Kernel'

Official Pypi Implementation of "SRI-Conv: Symmetric Rotation-Invariant Convolutional Kernel"

Yuexi Du, Nicha, Dvornek, John Onofrey

Yale University

This is the official release of SRI_Conv

version: 1.2.5

News

• [1.2.0] Gaussian Smooth Kernel is available now!
• [1.1.0] SRI_Conv1d and SRI_Conv3d are also available now.

Installation

pip install SRI-Conv

Requirement:

"scipy>=1.9.0",
"numpy>=1.22.0",
"torch>=1.8.0"

Note: Using lower version of torch and numpy should be fine given that we didn't use any new feature in the new torch version, but we do suggest you to follow the required dependencies. If you have to use the different version of torch/numpy, you may also try to install the package from source code at project repo.

Usage

>>> import torch
>>> from SRI_Conv import SRI_Conv2d, sri_resnet18
>>> x = torch.randn(2, 3, 32, 32)
>>> sri_conv = SRI_Conv2d(3, 16, 3)
>>> conv_out = sri_conv(x)
>>> sri_r18 = sri_resnet18()
>>> output = sri_r18(x)
# To reproduce the SRI-ResNet18 used in the paper, use:
>>> sri_r18 = sri_resnet18(ri_conv_size=[9, 9, 5, 5], skip_first_maxpool=True)