A messy attempt (in development) to implement the recurrent attention model (RAM) from "Recurrent Models of Visual Attention" (Mnih+ 2014)
###Dependencies
- TensorFlow
- matplotlib
###Execution
python ram.py
Initial run may take a minute, since it downloads MNIST
Visualization of glimpses is turned on by default
Training starts from iteration 10000 with an existing model by default