Towards Discriminative Latent Spaces for Compact Hashing and Image Retrieval

Pytorch code for reproducing the results of our group project for EEE598 course (Statistical Machine Learning: From Theory to Algorithms) taught by Profesor Gautam Dasarathy

Group members-

1.Kowshik Thopalli

2.Tejas Gokhale

The main codes are in the script folder. If you dont want to train, and just want to evaluate please use aae_pytorch_eval.py and change the checkpoint directory file to the desired prior ('gaussian' or'bernoulli' or 'disc_unif'). We have provided checkpoints for z_dim =10 for all of these three cases. This script will run a KNN classifier, plot the confusion matrix and also plots retrieved images for query test images.

Requirements-

PyTorch 1.0 with cuda

Torchvision

and usual Python ML libraries such as

Scipy, Numpy, pandas,

scikit-learn,

seaborn and matplotlib