Zero shot learning aims to solve a task to identify classes that are not seen by the model during training. For example, let’s say you have seen a horse but never seen a zebra. If I tell you that a zebra looks like a horse but it has black and white stripes, you will probably immediately recognize a zebra when you see it. This is what zero-shot learning aims to tackle.
Data has been collected from this website https://visualgenome.org/ 20 classes in total where there are 15 classes selected for training and 5 classes selected as Zero-Shot classes
arm, boy, bread, chicken, child, computer, ear, house, leg, sandwich, television, truck, vehicle, watch, woman
car, food, hand, man, neck
Dataset has been collected from this website https://www.mpi-inf.mpg.de/departments/computer-vision-and-machine-learning/research/zero-shot-learning/latent-embeddings-for-zero-shot-classification.
Different datasets with different length of attributes for each dataset
- SUN: Attributes 102, Classes 717
- CUB: Attributes 312, Classes 200
- AWA1: Attributes 85, Classes 50
- AWA2: Attributes 85, Classes 50
- APY: Attributes 64, Classes 32
Follow the below steps to run the code:
- Install all the necessary packages
- Open the jupyter notebook located at https://github.com/kiran74-ds/zsl/blob/main/Zero_Shot_Learning.ipynb
- Follow the steps in the note book
Follow the below steps to run the code:
- Install all the necessary packages
- Open the jupyter notebook located at https://github.com/kiran74-ds/zsl/blob/main/Embarssingly_Simple_ZSL.ipynb
- Follow the steps in the note book