https://app.diagrams.net/?mode=google&gfw=1#G1ffxccn-BOE_Bm5jI93tc76hmP6np0bAL

Diagram

Local Resources

To save on the high costs of cloud applies for persistent storage, we will use local resources. The servers below can be installed on a computer with fairly limited resources.

Miminum Recommended Local Computer:

OS: Windows 10 Pro
Prerequisites: Enable Hyper-V
CPU: i7 8700k or Higher
Memory: 32 GB
Disk: 1 TB

NextCloud – Free

We need a place to organize our files and processes. Nextcloud is an open source platform that offers an entire workspace solution similar to Microsoft 365. This is FREE! We can setup this with

Apache Guacamole – Free

We can manage our backend using a web client as well. Here we use open source application Guacamole. The setup does take some experience but we then have another free management platform.

Cloud Resources

Azure VM – Label-Studio – (~$0.60 per day)

Annotation is a crucial step in building a computer vision pipeline, and Label-Studio is a great tool to manage it. However, image annotation is a time-consuming task and you may not always need to use it. To keep costs low, we will be using an Azure VM. We have provisioned the disk for 30 GB as our dataset is small. With Azure VM Premium Disks, pricing can increase based on your provisioning needs, but the advantage is that you only incur charges when the service is in use. Additionally, by leaving the VM offline, we can further cut down our costs, which aligns with our goal of keeping expenses as low as possible.

On the server, I setup Label-studio inside a python virtual enviornment. You can find the same steps here. (Link)

Lambda Labs – Training Server – ($0.60/hour) Or Edge Impulse