flowerClassification

This project is my introductory project to the world of Deep Learning. I trained an image classifier to recognize different species of flowers. I used [Maria-Elena Nilsback and Andrew Zisserman's dataset of 102 flower categories](http://www.robots.ox.ac.uk/~vgg/data/flowers/102/index.html) for training. The flowers chosen to be flower commonly occuring in the United Kingdom. Each class consists of between 40 and 258 images.

I used ResNet as my pre-trained model for feature extraction and the following structure is of the classifier model layers on top of it

(classifier): Sequential(
    (fc1): Linear(in_features=9216, out_features=4096, bias=True)
    (relu1): ReLU()
    (dropout1): Dropout(p=0.3, inplace=False)
    (fc2): Linear(in_features=4096, out_features=2048, bias=True)
    (relu2): ReLU()
    (dropout2): Dropout(p=0.3, inplace=False)
    (fc3): Linear(in_features=2048, out_features=102, bias=True)
    (output): LogSoftmax()
  )

For list of all the classes, you can refer to cat_to_name.json file

Contribution to the project

Please follow a systematic Git Workflow -

• Create a fork of this repo.
• Clone your fork of your repo on your pc.
• Add Upstream to your clone
• Every change that you do, it has to be on a branch. Commits on master would directly be closed.
• Make sure that before you create a new branch for new changes,syncing with upstream is neccesary.

Commits

• Write clear meaningful git commit messages (Do read this).
• Make sure your PR's description contains GitHub's special keyword references that automatically close the related issue when the PR is merged. (Check this for more info)
• When you make very very minor changes to a PR of yours (like for example fixing a failing Travis build or some small style corrections or minor changes requested by reviewers) make sure you squash your commits afterward so that you don't have an absurd number of commits for a very small fix. (Learn how to squash at here)
• When you're submitting a PR for a UI-related issue, it would be really awesome if you add a screenshot of your change or a link to a deployment where it can be tested out along with your PR. It makes it very easy for the reviewers and you'll also get reviews quicker.

For those of you who want to use the trained model, download it from here. If you are interested to contribute to the model/project (see to-do below), please do drop me an email at [email protected] and we can work together on the training part.

Installation and Setup

Setup and running of project (Backend)

• Fork the repo and clone it.
• Go in the repo and setup virtualenvironment using python -m virtualenv env
• Then activate the environment using source env/Scripts/activate (Use only env/Scripts/activate if on cmd or powershell)
• For Linux Users it will be source env/bin/activate
• At the root of your project directory
  

pip install -r requirements.txt
pre-commit install

• This will setup the project requirements and pre-commit test hooks!
• Install Pytorch if you want to work on the model/training part.
• Start the backend server python backend/app.py

This runs the backend server at default port 5000. Open http://localhost:5000 to view it in the browser.

Setup and running of project (Frontend)

• At your root directory run yarn install to install all the dependencies
• To start react dev server yarn start

This runs the app in the development mode.
Open http://localhost:3000 to view it in the browser.

The page will reload if you make edits.

To-Do tasks

• Complete the frontend part [WIP]
• Add Indian Species of flowers
• Convert entire application to cross platfrom mobile app
• Display information about the predicted flower.
  • Blooming conditions
  • Typical diseases
  • Any significant properties
  • Favourable place/conditions
• Display scientific name of the specie as well
• Local name of the flower to be displayed