Deep Detect To Annotate Images With AI Models

[NeoMoritsiqu]

Hi Sebastian,

“Rules: Support for deep detect to annotate images with AI models.” in change logs. added. I downloaded the current version and examined it, but I could not find how to use this feature. Can you support me?

[Sebastian]

Sure, it is still pending to write documentation for that. DeepDetect is a server to run AI models and the integration is enables a rule to annotate images.

The reason for being a rule is mostly that deepdetect downloads the image from Squidex, because thats the only API.

To run it I recommend to use my docker images. it is basically the default image with the model preconfigured:

Either

• https://hub.docker.com/r/squidex/deepdetect_gpu
• https://hub.docker.com/r/squidex/deepdetect_cpu

Install the container and don’t make it public, there is no authentication. Then configure the URL in Squidex: https://github.com/Squidex/squidex/blob/master/backend/src/Squidex/appsettings.json#L681

Then you will have a deepdetect rule that annotates the image. I was also playing around with calculating the focus point automatically, but the results are not good enough.