Butterfly Classification AI Challenge

The aim of this challenge is to classify given images into 50 categories of butterflies.

link to challenge : https://dockship.io/challenges/5f7418d6fbaa7b0393ff1a05/butterfly-classification-ai-challenge/overview

This is an AI challenge by Doship where we have to train a deep learning model to classify 50 categories of butterfly. This is an image classification with small dataset which has about maximum of 120 image per class and minimum of 60.

Frist aproach

Here i used transfer learning for traning the model. Here I used Resnet50v2 and got an test accuracy of 81.2

second approach

Here i trained two pre trained model namely Xception and Resnet101v2 both has training accuracy of 83 and 86 respectively. I sum up the prediction probabilities of both model and applied argmax on the final result. So the test accuracy increased from 81.2 to 89.400

Final approach