Zero-shot Image Recognition: Image classification

Academic Project (CS244-Introduction to data science)

zero-shot-learning

Implementation of Zero-Shot Learning algorithm

Zero-Shot learning method aims to solve a task without receiving any example of that task at training phase.
It simply allows us to recognize objects we have not seen before.

Classes

Train Classes:
arm, boy, bread, chicken, child, computer, ear, house, leg, sandwich, television, truck, vehicle, watch, woman
Zero-Shot Classes:
car, food, hand, man, neck

Usage

$ python3 detect_object.py input-image-path

Example

$ cd src
$ python3 detect_object.py ../test.jpg

--- Top-5 Prediction ---
1- vehicle
2- truck
3- car
4- house
5- watch

Test image is a beautiful Jaguar.
All related prediction results are ranked in first three.

P.S. Remember, the prediction results are only allowed to be among above classes (train and zero-shot classes).
Algorithm will fail (although it will do its best to predict most related class) in case you try to detect an object from different other classes.

Model Summary

ZSL Model Summary