A project supervised by Dr. Adrian Bors, titled "Computer Learning of what is Memorable from Images", where I explored the use of CNNs to predict the most and least memorable regions of images.
I explored several deep learning techniques and settled on Transfer Learning, which I implemented in PyTorch.
This project contained Visual Memory Schema (VMS) like the following, which show the most memorable regions of images.
I attempted to train a CNN to generate such VMS, with disappointing results, but the experience I gained about Transfer Learning and the deep knowledge of PyTorch and general Python while doing this project was more valuable than any results I could've hoped for.
A sample of "good" VMS maps I managed to generate are shown below (Top are generated, bottom are actual VMS maps from the dataset).
