Can a computer see similarity like a human?

In this project, we train a Siamese Neural Network to identify the odd-one-out image in a set of triplet images from the THINGS Similarity dataset of the THINGS database.

The inspiration for this project comes from a NIH paper that demonstrates how human object similarity can be recreated through an embedding matrix and a modeling algorithm. Our implementation strives to achieve the same objective but leverages a neural network instead.

To view our work, click on the NMA_Project.ipynb file above.

By: You, Bonnie, Bereket, Melika