Dog Breed Classification Project

Project Functionalities

• Users can know more about dog breeds and their information through taking photos or uploading photos.
• Users can also directly use photos provided by the project to quickly play around.
• Users can get top 3 dog breed prediction results with wikipedia links.
• Users can learn English and Chinese at the same time.

System Architecture

• dog detection
  • model: MobileNet V2 pretrained model without finetuning
  • input: an image having RGB channels with Base64 encoding
  • output: boolean
• dog classification
  • model: MobileNet V2 pretrained model with finetuning
  • input: an image having RGB channels with Base64 encoding
  • output: top-3 prediction results(dog breed names and prediction probabilities)

Project Structure

• api: backend
• web: frontend

Used Tools/Technologies

• frontend
  • HTML/CSS/JavaScript
  • Materialize
  • Babel
  • Webpack
• backend
  • Python
    • FastAPI
  • Docker
  • GCP
    • Cloud Run
• deep learning
  • PyTorch

Deep Learning Under the Hood in This Project

You can check this repo for further details, and this project is orginally from my Udacity Machine Learning Engineer Nanodegree capstone project. However, I have changed some model details for this dog breed classifier, you can check here for more details.

Setup

Frontend

please see the web folder

• npm install (download dependencies first)
• local development
  • npm run start
• deployment
  • npm run build

You can checkout other commands in package.json in the web folder

Backend

please see the api folder

• local development without docker:
  • pip install pipenv (If you don't have pipenv installed)
  • pipenv install
  • pipenv run uvicorn main:app --reload

Below commands are already written in Makefile, please checkout and change any setting you need first.

• local development using docker:
  • make dev
• build a docker image:
  • make build
• submit a docker image to Google Container Registry:
  • make submit
• deploy a docker image in Google Container Registry to Google Cloud Run:
  • make deploy

Todos

• support English
• support PWA
  • Android
  • iOS
• export the trained PyTorch models to edge devices for better inference performance