AnnoMachine

What is AnnoMachine?

AnnoMachine is an application that aims to provide a user-friendly way to visualize and interact with object detection results.

AnnoMachine fulfills that mission by:

• Running a deep learning model under the hood to get a rough estimation of the annotation

• Providing a user-friendly interface which user can easily visualize and interactively update the annotation

Development Tools:

Stack:

• Object detection model: SSD (implemented on Tensorflow 2)
• Backend: Flask
• Frontend: ReactJS (SVG for displaying image and drawing bounding boxes)
• Database: PostgreSQL
• Web server: Nginx

Unit test:

• Backend: Flask-Testing
• Frontend: Jest + Enzyme

Deployment:

• docker
• docker-compose
• docker-machine

Preparation

Clone the repo:

git clone https://github.chunml/annomachine
cd annomachine

Download the pretrained weights (trained on VOC07+12 for 120 epochs): weights

Save the weights to ./services/backend/ssd_tf2/models:

mkdir -p ./services/backend/ssd_tf2/models
mv ssd_epoch_120.h5 ./services/backend/ssd_tf2/models

Make sure the unit tests are passed:

./test.sh

Local deployment

What we need to do is to run the command below:

docker-compose up -d --build

Head to localhost and see the result!

AWS deployment

• Create your own Access Key ID and Secret Access Key from AWS
• Create a new file called ~/.aws/credentials (a file, not a folder)
• Paste the below content (replace with your keys):

[default]
aws_access_key_id=<your_access_key_id>
aws_secret_access_key=<your_secret_access_key>

• Create a new machine via docker-machine

docker-machine create --driver amazonec2 --amazonec2-open-port 80 <your_machine_name>

This will take a while. After that, a new EC2 instance is created on AWS, using the credentials you created above.

• Check if the new machine is created properly:

docker-machine ls

• Activate the AWS machine:

eval $(docker-machine env <your_machine_name>)

It means that from now on, we are inside the environment of the AWS machine! What we're gonna see/do all happens within that environment.

• Check if we are actually in AWS machine's environment:

docker-machine ls

We should see an asterisk next to the machine's name. Let's take note the IP address.

• Make sure that there is nothing on there yet:

docker image ls

• Deploy onto the AWS machine

docker-compose -f docker-compose-prod.yml up -d --build

• Confirm that the deployment is successful: Head to <your_machine_IP> and enjoy the result!