Code and supplements for the article "Visual Attention Through Uncertainty Minimization in Recurrent Generative Models"
All relevant code can be found in the directory. The model is specified in <model.py> and <modules.py>. The training and testing loops can be found in <trainer.py>. <data_loader.py> and <config.py> handle the data and configuration of all simulations. The configuration files used to train the models in the articles can be found in the folder.
calls the test runs with the pretrained models in the dir and displays the figures shown in the article. To run the notebook the MNIST, translated MNIST and, cluttered MNIST data have to be generated using https://github.com/deepmind/mnist-cluttered and placed into a folder.
All required python packages should be found in <requirements.txt>.