🌼 Flower Classifier

This project leverages deep learning algorithms to classify different kinds of flowers with high accuracy.

This project assumes familiarity with Pytorch, Pytorch Lightning, Weights and Biases (W&B), and Gradio, but some basic commands are available to get you started.

Prerequisite

This project uses W&B to keep track of experiments. Thus, in order to proceed further steps, a W&B account is required.

Setting up the Python environment

pipenv install --ignore-pipfile

Using the pretrained model

python training/stage_model.py --fetch --entity=khoale --from_project=flower_classification

Fine-tuning a deep learning model on the flower dataset

python training/run_experiment.py --max_epochs=8 --gpus='0,' \
--num_workers=24 --model_class=VGG16Classifier --data_class=Flowers \
--fc1_dim=8192 --fc2_dim=2048 --batch_size=32 --wandb

Feel free to change the values of fc1_dim, fc2_dim, or to get rid of the --gpus flag if you don't have ones.

Serializing the trained model

python training/stage_model.py --entity='your_account_name'

Running the gradio app

Approach 1:

If the pretrained model has been downloaded and lived in the right directory, you can run the following command to enjoy the final product.

python flower_classifier/app_gradio/app.py

Approach 2:

Build a container image and get the container up and running, using the following commands:

docker build -t flower-model-backend:1.0.0 . --file api_serverless/Dockerfile
docker run -p 9000:8080 -d flower-model-backend:1.0.0

Now that the model server has been up and running, the Gradio app can be started.

python app_gradio/app.py --model_url=http://localhost:9000/2015-03-31/functions/function/invocations

Deploying to AWS

• Build a container image from the provided Dockerfile
• Upload the container image to the Elastic Container Registry (ECR)
• Create a Lambda function
• Add an HTTP endpoint with a URL
• Connect to the frontend