Dog Breed Classification 🐶

A demo that uses a ResNet model to predict the breed of a dog given in a photo.

Link to our blog post https://www.data-max.io/post/dog-breed-classification.

Geting Started

There are two ways you can run this project:

• Docker
• Kubernetes

Start by cloning the repository

git clone https://github.com/data-max-hq/dog-breed-classification-ml.git

Run project with docker:

Prerequisites:

• Docker
• Docker Compose

1. Open the project directory in terminal and type

   make requirements

2. Train the model (only once)

   make local-train

3. Deploy model using:

• TensorFlow Serving

  make compose-tfserve

• Seldon Serving

  make compose-seldon

4. Open Streamlit UI at http://localhost:8502. Enjoy predicting 🪄

5. Stop docker containers

   docker compose down

Run project with Kubernetes:

Prerequisites:

• Docker
• Helm
• Helmfile
• Minikube

1. Deploy model using:

• TensorFlow Serving
  1. Create a kubernetes cluster (minikube)

     make start-tfserve

  2. Build images

     make build-tfserve

  3. Load images to minikube

     make load-tfserve

  4. Install Kubeflow

     make install-kubeflow
     #Wait till all pods of kubeflow are running

  5. Expose Kubeflow port so you can access Kubeflow dashboard at http://localhost:8080 (optional)

     make port-kubeflow

  6. Deploy TensorFlow Serving, Ambassador and Streamlit

     make helm-tfserve

  7. Apply mapping resources

     make deploy-tfserve

  8. Expose Emissary-ingress port

     make port-emissary

• Seldon Serving
  1. Create a kubernetes cluster (minikube)

  make start-seldon

  2. Build images

     make build-seldon

  3. Load images to minikube

     make load-seldon

  4. Install Kubeflow

     make install-kubeflow
     #Wait till all pods of kubeflow are running

  5. Expose Kubeflow port so you can access Kubeflow dashboard at http://localhost:8080 (optional)

     make port-kubeflow

  6. Deploy Seldon-Core, Ambassador and Streamlit

     make helm-seldon

  7. Deploy Seldon application and apply mapping resources

     make deploy-seldon

  8. Expose Emissary-ingress port

     make port

2. Open Streamlit UI at http://localhost:8080/streamlit. Enjoy predicting 🪄
3. Delete cluster

   make delete