DogBreedClassifier

Model used for the final project of TI2022 - Dog Breed classifier app for local Veterinary

A model for classifying 120 Dog Breeds built using the EfficientNet-B3 feature vector pre-trained on Imagenet (ILSVRC-2012-CLS) and fine-tuned for the task at hand. The model achieves an overall accuracy of 86.7%

Getting started 🚀

To get a working environment there are two possible options.

1. Create a conda environment with the listed pre-requisites
2. Create a conda environment from the .yml file

Pre-requisites 📋

Software and dependencies needed

pytorch-1.7.0
tqdm-4.54.1
torchvision-0.8.1
numpy-1.19.2

Installation 🔧

To get started make sure you either have the listed pre-requisites or set up the anaconda environment from the .yml file.

conda env create -f environment.yml

Make sure you activate the environment.

conda activate DogNet

And verify that it was properly installed.

conda env list

Data

The dataset used to train the model was the Stanford Dogs Dataset which contains 20,580 images with a total of 120 dog breeds.

Acknowledgments 🎁

Special thanks to Aladding Persson aladdinpersson for his tutorial on Building a Dog Breed Identifier App from scratch - DogNet