jbednar
commented
9 years ago Yes. For many years, Topographica has included a LISSOM-based motion map model in models/lissom.ty, which can be used to replicate the results from http://nn.cs.utexas.edu/computationalmaps/figures/5.23.php,
http://nn.cs.utexas.edu/keyword?bednar:neurocomputing03, and
http://nn.cs.utexas.edu/keyword?bednar:neurocomputing06. However, that .ty file is very difficult to use and understand, we no longer use LISSOM for current models, and the model of motion used there has only a tenuous relationship to real chronological time.
In the latest Github version of the source code, we now have a GCAL-based motion model in examples/gcal.ty, which be enabled by passing "dims=['xy','or','dr']" to the call to ModelGCAL() in that file. An example IPython notebook showing how to use and analyze this network is at
http://ioam.github.io/media/topo/gcal_all.html, though motion is not enabled by default, so you'll need to re-run the notebook doc/Tutorials/gcal_all.ipynb after changing simulations["fischer:ms14 fig6.7-8"] to simulations["fischer:ms14 fig6.9-11"]. This model has not been analyzed in detail, but it does have GCAL's robustness and simplicity, and has a much simpler and clearer relationship to chronological time. Thus the GCAL-based model is a better starting point, even though it's very much still a work in progress.
Hi, Has anybody used Topographica for V1-like directional motion detection?