scatnetgpu 1.0

Scattering Network for Python and CUDA

Description

scatnetgpu is a module to compute the Scattering Network representation [1] of an image using the power of a CUDA capable GPU.

The output of the computation is compatible with the software scatnet.

Requirements

To run the main code:

• Python 2.7 (not yet tested with Python 3.x)

• A CUDA capable GPU with CUDA toolkit correctly installed

To create filters:

• Octave installed

• scatnet installed and in the default octave path

Installation

Just run:

$ pip install scatnetgpu

Alternatively, clone the repository and run:

$ pip install .

Quickstart

Load or create an image as a Numpy’s ndarray. Here we load an image with OpenCV:

import cv2
img = cv2.imread('image.png')

Create the ScatNet object:

from scatnetgpu import ScatNet
sn = ScatNet(M=2, J=4, L=6)

Perform the actual transformation

out = sn.transform(img)

Now out contains the features of the Scattering Network representation of img.

References

[1] Bruna, Joan, and Stéphane Mallat. “Invariant scattering convolution networks.” IEEE transactions on pattern analysis and machine intelligence 35.8 (2013): 1872-1886.