Applied Machine Intelligence Project

It is the repository of the final project for Applied Machine Intelligence (SS23) at Technical University Munich.

Team Member:

• Huixu Liu (Responsible for: Deployment)
• Lei Zhang (Responsible for: Active Learning)
• Linghui Liu (Responsible for: Web Inteface)
• Mengqi Wei (Responsible for: Calssification(DL-based Methods))
• Yinjia Wang (Responsible for: Classification(ML-based Methods), Web Interface)
• Zihan Xu (Responsible for: Segmentation, Web Interface)

Introduction

This project aims to help clinicians to identify blood cell types. The main functions are segmentation, classification and active learning.

Requirements

This project requires Python 3.9 or higher and uses the following Python packages:

• pytorch
• torchvision
• scikit-learn
• numpy
• matplotlib
• django
• openCV
• tqdm
• h5py

User's Guide

💻 : work for Windows, Linux and macOS

To install the necessary packages, use pip:

pip install torch
pip install torchvision
pip install scikit-learn
pip install numpy
pip install matlabplot
pip install django
pip install opencv-python-headless
pip install tqdm
pip install h5py

Remote Version

The latest image is already available in the container registry and can be deployed to the remote k8s using flux.

First you need to:

• Getting the kubectl and flux clients

• Make sure you're in MWN (accessed from home using EduVPN)

• Set the KUBECONFIG environment variable to the configuration file kubeconfig_group02.yaml in the repository.

  1.install the flux tool locally https://fluxcd.io/flux/installation/

  2.Use this command to set the environment variable

  $ export KUBECONFIG=kubeconfig_group02_group02.yaml

  This gives you operational access to the remote k8s.

  3.Then use the command line

  $ flux reconcile kustomization -n group02 group02 --with-source

  to configure the files under /deployment to the remote k8s.

  4.Use the URL 'https://cellai.group02.ami.dedyn.io/' to access the page opened on the remote.

Local Version

Reset to commit 232cfc46. Copy the 'inputs' and 'outputs' folders in ami project group share group02/model and replace the folders with the same names in the cloned repository Group02/app/web/src/model.

To run the Django server, switch to Group02/app/web/src, use the following command:

python manage.py runserver

Then open a web browser and navigate to http://127.0.0.1:8000/ to see the website in action.

In the homepage, upload a h5py file and wait for the prediction.

In the results gallery, click an image and modify the prediction label. Click UPDATE & TRAINING to retrain the model and predict.