CoordConv for Keras

Keras implementation of CoordConv from the paper An intriguing failing of convolutional neural networks and the CoordConv solution.

Extends the

CoordinateChannel

Usage

Import

coord.py

There are 3 different versions of CoordinateChannel - 1D, 2D and 3D for each of

Conv1D

Conv2D

Conv3D

from coord import CoordinateChannel2D

prior to first conv
ip = Input(shape=(64, 64, 2))
x = CoordinateChannel2D()(ip)
x = Conv2D(...)(x)  # This defines the CoordConv from the paper.
...
x = CoordinateChannel2D(use_radius=True)(x)
x = Conv2D(...)(x)  # This adds the 3rd channel for the radius.

Experiments

The experiments folder contains the

Classification

Uniform

Quadrant

Creating the datasets

First, edit the

make_dataset.py

type

uniform

quadrant

Uniform Dataset

The uniform dataset model can be trained and evaluated in less than 10 epochs using

train_uniform_classifier.py

|Train | Test | Predictions | |:---: | :---: | :-----------: | | | | |

Quadrant Dataset

The quadrant dataset model can be trained and evaluated in less than 25 epochs using

train_quadrant_classifier.py

|Train | Test | Predictions | |:---: | :---: | :-----------: | | | | |

Checks

To see if the implementation of CoordConv index concatenation is correct, please refer to the numpy implementations in the

checks

Difference from paper

This implementation of the coordinate channels creation differs slightly from the original paper.

The major difference is that for 2/3D Convolutions, it may not be the case that the height and width are the same for all layers. The original implementation would throw an error due to shape mismatch during the concatenation.

To over come this, the

np.ones()

This modification along with some transpose operations allows for height and width to be different and still work.

Theano Support

Theano is partially supported with the

coord_theano.py

Requirements

• Keras 2.2.0+
• Either Tensorflow or CNTK backend.
• Matplotlib (to plot images only)