Skip to content

Latest commit

 

History

History
77 lines (54 loc) · 2.67 KB

README.md

File metadata and controls

77 lines (54 loc) · 2.67 KB
Raw

Azure ML

Overview

This repository utilizes Azure ML to train detectron2 and yolact based models. Refer to the below diagram for an architecture overview of the implemented method.

AML Demo Development Status and Content

Azure Machine Learning Service Demo

Quick Start

Quick start for utilized detectron2 and yolact for AzureML training. Please follow this Azure ML Doc for setting up initial AzureML Environment.

  1. Finalize training dataset and coco annotations
  2. update model training configs and dataset info in either the aml_detectron2 or aml_yolact aml_code folder
  3. update required training config in the experiment_cfg.py file
  4. run python aml_new_experiment.py

Installation

  1. Clone this repository (git installation required)

    cd $HOME # or another directory for this project/repo
git clone https://github.com/naivelogic/azureml_cv.git
cd azureml_cv

  2. Tools that I am using

  3. Install environment with Anaconda:

    conda env create -f azureml-env.yml 
conda activate azureml

## if using Jupyter Notebooks create custom jupyter kernel for AzureML
python -m ipykernel install --user --name=azureml

jupyter nbextension install --py --user azureml.widgets
jupyter nbextension enable azureml.widgets --user --py

  4. Login to Azure on the command line

    seems like i need to run az login --use-device-code to initialize each session for authentication to the AML workspace (i have multiple AML across different subscriptions and tenants)

    # azure cli linux 
## install az function tools
# https://docs.microsoft.com/en-us/cli/azure/install-azure-cli-linux?pivots=apt
curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash

# now log in 
az login --use-device-code

# may also need to install this package if getting authentication errors
pip install azure-cli-core