Dog breed classifier using CNN and Transfer Learning

Project Overview

In this project, the code will accept any user-supplied image as input. If a dog is detected in the image, it will provide an estimate of the dog's breed. If a human is detected, it will provide an estimate of the dog breed that is most resembling.

Import Datasets

• Download the dog dataset
• Download the human_dataset

Create a CNN to Classify Dog Breeds(Building a model on my own)

Result

Hi, This is a dog!
And I predict the Breed of this dog is a ages/train/076.Golden_retriever

Summary

There are several state of the arts model we tried. First I gave to shot to OpenCv but it fails when we don't have forward facing faces. Then I tried convolutional Neural Network from scratch with above parameters.It gave an accuracy of 9% which is too low. At last i tried two state of the arts pre-trained model VGG16 and Xception. VGG16 gave an accuracy of 43% where as VGG19 gave an accuracy of 70.9% of validation dataset with 20 epochs and about 30 sec of training on GPU.