e2cnn is a PyTorch extension for equivariant deep learning.
Equivariant neural networks guarantee a specified transformation behavior of their feature spaces under transformations of their input.
For instance, classical convolutional neural networks (CNNs) are by design equivariant to translations of their input.
This means that a translation of an image leads to a corresponding translation of the network's feature maps.
This package provides implementations of neural network modules which are equivariant under all isometries E(2) of the image plane
, that is, under translations, rotations and reflections.
In contrast to conventional CNNs, E(2)-equivariant models are guaranteed to generalize over such transformations, and are therefore more data efficient.
The feature spaces of E(2)-Equivariant Steerable CNNs are defined as spaces of feature fields, being characterized by their transformation law under rotations and reflections. Typical examples are scalar fields (e.g. gray-scale images or temperature fields) or vector fields (e.g. optical flow or electromagnetic fields).
Instead of a number of channels, the user has to specify the field types and their multiplicities in order to define a feature space.
Given a specified input- and output feature space, our R2conv module instantiates the most general convolutional mapping between them.
Our library provides many other equivariant operations to process feature fields, including nonlinearities, mappings to produce invariant features, batch normalization and dropout.
Feature fields are represented by GeometricTensor objects, which wrap a torch.Tensor with the corresponding transformation law.
All equivariant operations perform a dynamic type-checking in order to guarantee a geometrically sound processing of the feature fields.
E(2)-Equivariant Steerable CNNs unify and generalize a wide range of isometry equivariant CNNs in one single framework. Examples include:
- Group Equivariant Convolutional Networks
- Harmonic Networks: Deep Translation and Rotation Equivariance
- Steerable CNNs
- Rotation equivariant vector field networks
- Learning Steerable Filters for Rotation Equivariant CNNs
- HexaConv
- Roto-Translation Covariant Convolutional Networks for Medical Image Analysis
For more details we refer to our NeurIPS 2019 paper General E(2)-Equivariant Steerable CNNs.
The library is structured into four subpackages with different high-level features:
| Component | Description |
|---|---|
| e2cnn.group | implements basic concepts of group and representation theory |
| e2cnn.kernels | solves for spaces of equivariant convolution kernels |
| e2cnn.gspaces | defines the image plane and its symmetries |
| e2cnn.nn | contains equivariant modules to build deep neural networks |
Since E(2)-steerable CNNs are equivariant under rotations and reflections, their inference is independent from the choice of image orientation. The visualization below demonstrates this claim by feeding rotated images into a randomly initialized E(2)-steerable CNN (left). The middle plot shows the equivariant transformation of a feature space, consisting of one scalar field (color-coded) and one vector field (arrows), after a few layers. In the right plot we transform the feature space into a comoving reference frame by rotating the response fields back (stabilized view).
The invariance of the features in the comoving frame validates the rotational equivariance of E(2)-steerable CNNs empirically. Note that the fluctuations of responses are discretization artifacts due to the sampling of the image on a pixel grid, which does not allow for exact continuous rotations.
For comparison, we show a feature map response of a conventional CNN for different image orientations below.
Since conventional CNNs are not equivariant under rotations, the response varies randomly with the image orientation. This prevents CNNs from automatically generalizing learned patterns between different reference frames.
E(2)-steerable convolutions can be used as a drop in replacement for the conventional convolutions used in CNNs. Keeping the same training setup and without performing hyperparameter tuning, this leads to significant performance boosts compared to CNN baselines (values are test errors in percent):
| model | CIFAR-10 | CIFAR-100 | STL-10 |
|---|---|---|---|
| CNN baseline | 2.6 ± 0.1 | 17.1 ± 0.3 | 12.74 ± 0.23 |
| E(2)-CNN * | 2.39 ± 0.11 | 15.55 ± 0.13 | 10.57 ± 0.70 |
| E(2)-CNN | 2.05 ± 0.03 | 14.30 ± 0.09 | 9.80 ± 0.40 |
The models without * are for a fair comparison designed such that the number of parameters of the baseline is approximately preserved while models with * preserve the number of channels, and hence compute. For more details we refer to our paper.
e2cnn is easy to use since it provides a high level user interface which abstracts most intricacies of group and representation theory away. The following code snippet shows how to perform an equivariant convolution from an RGB-image to 10 regular feature fields (corresponding to a group convolution).
1| from e2cnn import gspaces
2| from e2cnn import nn
3| import torch
4|
5| r2_act = gspaces.Rot2dOnR2(N=8)
6| feat_type_in = nn.FieldType(r2_act, 3*[r2_act.trivial_repr])
7| feat_type_out = nn.FieldType(r2_act, 10*[r2_act.regular_repr])
8|
9| conv = nn.R2Conv(feat_type_in, feat_type_out, kernel_size=5)
10| relu = nn.ReLU(feat_type_out)
11|
12| x = torch.randn(16, 3, 32, 32)
13| x = nn.GeometricTensor(x, feat_type_in)
14|
15| y = relu(conv(x))Line 5 specifies the symmetry group action on the image plane
under which the network should be equivariant.
We choose the
cyclic group
C8, which describes discrete rotations by multiples of 2π/8.
Line 6 specifies the input feature field types.
The three color channels of an RGB image are thereby to be identified as three independent scalar fields, which transform under the
trivial representation
of C8.
Similarly, the output feature space is in line 7 specified to consist of 10 feature fields which transform under the
regular representation
of C8.
The C8-equivariant convolution is then instantiated by passing the input and output type as well as the kernel size to the constructor (line 9).
Line 10 instantiates an equivariant ReLU nonlinearity which will operate on the output field and is therefore passed the output field type.
Lines 12 and 13 generate a random minibatch of RGB images and wrap them into a nn.GeometricTensor to associate them with their correct field type.
The equivariant modules process the geometric tensor in line 15.
Each module is thereby checking whether the geometric tensor passed to them satisfies the expected transformation law.
A hands-on tutorial, introducing the basic functionality of e2cnn, is provided in introduction.ipynb. Code for training and evaluating a simple model on the rotated MNIST dataset is given in model.ipynb.
More complex equivariant Wide Resnet models are implemented in e2wrn.py. To try a model which is equivariant under reflections call:
cd examples
python e2wrn.py
A version of the same model which is simultaneously equivariant under reflections and rotations of angles multiple of 90 degrees can be run via:
python e2wrn.py --rot90
torch>=1.1
numpy
scipy
Optional:
pymanopt
autograd
pip install .
The development of this library was part of the work done for our paper General E(2)-Equivariant Steerable CNNs. Please cite this work if you use our code:
@inproceedings{e2cnn,
title={{General E(2)-Equivariant Steerable CNNs}},
author={Weiler, Maurice and Cesa, Gabriele},
booktitle={Conference on Neural Information Processing Systems (NeurIPS)},
year={2019},
}
Feel free to contact us.
e2cnn is distributed under BSD Clear license. See LICENSE file.