Leveraging spiking deep neural networks to understand the neural mechanisms underlying selective attention
by Lynn K.A. Sörensen, Davide Zambrano, Heleen A. Slagter, Sander M. Bohté, & H. Steven Scholte
This is the code that accompagnies this paper and consists of three parts:
- the
asn
package for DCNN to sDCNN conversion with and without spatial attention - a set of function (
Datasets
) to replicate the dataset curation from the COCO database - the code to reproduce the results in the paper (
ModelTraining
,ModelEvaluation
,ModelAnalysis
) - the code to reproduce the paper figures (
Figures
)
The asn
package relies on Keras with a TensorFlow backend.
The dataset curation
relies on COCO API
as well as Deep Gaze II
.
For the analysis part, results files can be downloaded here to follow these analyses.
Please make sure to add the right files to the ModelEvaluation
and ModelAnalysis
folder to reproduce the Figures
.
The model training scripts can be found in ModelTraining
. The resulting weights are provided here.
Last updated: 10.11.2021