arifoyong / dog_breed_identifier

Dog Breed Identification

A simple project to predict dog breeds. This projects employed a deep learning model that has been trained to recognize dog breeds. The model is served through REST API running on python backend. A simple webapp is employed as user interface.

Demo note: performance is slow as we are running on cheap web server with shared other applications

Dataset & Model

The model was trained with dataset obtained from Kaggle. This dataset pictures of 120 different breeds of dogs.

Inceptionv3 is used as the base architecture. The model was developed through transfer learning from a pre-trained model trained on ImageNet dataset. During test the model was able to achieve 80.80% accuracy.

The training notebook is included in notebook folder.

This dataset comprises 120 breeds of dogs.

Resnet50 is used as the base architecture. The model was developed through transfer learning from a pre-trained model trained on ImageNet dataset. During test the model was able to achieve 89.1% accuracy.

How to use

Clone the repository to your local folder. Type the following:

docker-compose up

Go to: http://localhost:3001/

By default, backend server is serving the API on http://localhost:8000 If the default is changed, we have to let the frontend know. This can be done by creating .env file in /frontend and specifying API endpoint as follow:

API=new_API_endpoint

Technology used

Machine Learning:

1. Tensorflow

Backend:

1. Pytorch
2. FastAI

Backend:

1. Python
2. FastAPI
3. Docker

Frontend

1. Next.js
2. Tailwindcss
3. Docker