Zero shot learning is the problem of classification when the test data contains data from classes that have not been seen during training time. This project extends the work of Verma and Rai, a simple generative framework, which is based on estimating class-attributed-gated class condition distributions.

The basic file structure for the output is : <real_class> : <top 5 predicted classes> : <probability of top 5 classes>

-Abhibhav Garg, Prannay Khosla, Soumye Singhal

-Piyush Rai