cookiecutter-panorama

Cookiecutter template for AWS Panorama projects.

• Gentle introduction: Deploy your first computer vision app for AWS Panorama on Towards Data Science
• GitHub repo: https://github.com/mrtj/cookiecutter-panorama/
• Free software: MIT license

Features

• Easy-to-use Makefile based build system with rich features
• Test Utility integrated
• Git hooks to prevent leaking your AWS account identifier
• Batteries included: if you do not have a ready deep learning model, a dummy pytorch model is created for you so you can build and deploy the whole application right away.

Quick start

Install the latest Cookiecutter if you haven't installed it yet (this requires Cookiecutter 1.4.0 or higher):

pip install -U cookiecutter

Generate a Panorama application project:

cookiecutter https://github.com/mrtj/cookiecutter-panorama.git

Then:

• Initialize a git repository in your new project with make init-repo.
• Install the required build tools: aws-cli, panorama-cli and docker. Configure the aws-cli to the credentials of the account where the Panorama appliance is registered.
• Import the project to your AWS account with make import
• Build the project with make build. This will also create a dummy deep learning model that you can replace later with a real model. The dummy model will simply calculate the mean of the RGB channels of the input video frame.
• Upload the compiled application container and packaged deep learning model with make package. This script will also output the path of the compile manifest json.
• Pick up the manifest json, head to AWS Panorama Console and deploy your application!

Refer to the README.md of the generated project for more information.

Project template parameter reference

parameter	default value	description
project_name	Panorama Video Processor	The human-readable name of the project
project_slug	panorama_video_processor	Project slug that can be used in file names, identifiers, etc. Should contain only letters, numbers and underscores.
s3_working_bucket	my_bucket	An AWS S3 bucket where the build system have read/write privileges.
s3_working_path	s3://my_bucket/­panorama_projects/­panorama_video_processor	A full S3 URI in the bucket above. This path will be used for model compilation, archiving manifest files and similar.
camera_node_name	camera_input	The name of the camera input node in the manifest file. In most of the cases you can leave it to the default value.
display_node_name	display_output	The name of the display output node in the manifest file. In most of the cases you can leave it to the default value.
code_package_name	panorama_video_processor_logic	The name of the code package. The package name - package version tuple should be unique in your AWS account.
code_package_version	1.0	The code package version. You can have several versions of your package deployed contemporarily to your account.
code_asset_name	panorama_video_processor_logic_asset	The code asset name.
code_node_name	panorama_video_processor_logic_node	The name of the code node in the manifest file.
model_package_name	panorama_video_processor_model	The name of the model package. The package name - package version tuple should be unique in your AWS account.
model_package_version	1.0	The model package version. You can have several versions of your package deployed contemporarily to your account.
model_asset_name	panorama_video_processor_model_asset	The model asset name.
model_node_name	panorama_video_processor_model_node	The name of the model node in the manifest file.
model_input_name	input0	The name of the input of the deep learning model. The model will be compiled with this input name, and you should refer the input with this name from the application source code.
model_batch_size	1	The inference batch size of the deep learning model. You can process this number of images simultaneously with the model.
model_processing_width	224	The width of the input image of the deep learning model. The input frame will be resized to this size before sending it to the model.
model_processing_height	224	The height of the input image of the deep learning model. The input frame will be resized to this size before sending it to the model.