Official package for 'Symmetric Rotation-Invariant Convolutional Kernel'
Project description
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
andSRI_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)
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
sri_conv-1.2.5.tar.gz
(13.8 kB
view hashes)
Built Distribution
SRI_Conv-1.2.5-py3-none-any.whl
(14.6 kB
view hashes)