Most of my code is on my GitHub repository.
HMax is a biologically inspired model of visual recognition, developed mostly by Tomaso Poggio, Maximilian Riesenhuber and Thomas Serre. I reimplemented it in about 200 lines of Python code (and about as many lines of comments). It requires a working Python interpreter, with the Numpy and Scipy libraries installed. The code is linked to at the bottom of this page.
The whole model takes about a minute to run on a single 256x256 image. You might find it useful if you want to study or expand HMax. Otherwise, there is no real reason to use this. In particular, for any computational modeling task, you would be much better off using Deep Learning frameworks like Keras. (See also Instant Espresso from Chuan-Yung Tsai at the Cox lab, which leverages the Caffe deep learning architecture with just one page of very fast Python code.)
For more information on HMax:
- A feedforward architecture accounts for rapid categorization. T. Serre, A. Oliva, T. Poggio, PNAS 104:6424-6429, 2007. Also be sure to check the supplementary materials on Thomas Serre's website.
- Learning a dictionary of shape-components in visual cortex: Comparison with neurons, humans and machines. T. Serre, PhD Thesis, MIT, 2006
- Toward a More Biologically Plausible Model of Object Recognition. M. Kouh, PhdThesis, MIT, 2007
- A theory of object recognition: computations and circuits in the feedforward path of the ventral stream in primate visual cortex. T. Serre, M. Kouh, C. Cadieu, U. Knoblich, G. Kreiman, T. Poggio. MIT CSAIL Technical Report AI-2005-036, 2005.
During my PhD I worked on building simulations of 3D articulated robots ("virtual creatures"), controlled by neural networks, and designed by evolutionary methods. See Shane Celis' Github repository for a refactored, extended version of my original code.