AIQuS manual and code

[beltramoeinventio]

Summary

Say what this feature is about in a couple of sentences. Keep technical details out, focus on functionality. Have a look at the issues in GitHub roadmap for examples of writing this kind of thing.

Intended outcome

What's the end goal? What are the benefits? Again, have a look at the the issues in GitHub roadmap for inspiration.

How will it work?

Sketch out the high-level flow, keeping out as many technical details as possible.

Where we can find the code, installation procedure and communication specifications about AIQuS package? We have to integrate it in our solution, but at moment we don't find material on repository except this page: https://github.com/c0c0n3/kitt4sme/blob/master/arch/catalogue/aiqus.md

Thank you

[RyanKelvinFord]

Please provide your contact details so we can email you the requested information.

[beltramoeinventio]

You can send information to: info@inventio.ai Thank you very much

[beltramoeinventio]

We have not yet found information in the provided email, has it already been sent? Thank you

[beltramoeinventio]

Thant you for information and SW. Honewer during test of AIQUS tool we fall in follow drawbacks:

• The use of a shared folder should be problematic in a distribuited system, when the request of augmentation is sent, there aren’t feedback of when the images are really available in shared folder, so we had to make a constant polling of shared folder to verify when the images are completed
• If the AIQUS fails to elaborate the images (i.e due a corrupted file), there aren’t feedback of error
• The tool increase incrementally the memory usage after repeated requests. In some case it reach >5 Gb of RAM usage. This should be an issue in a production environment.
• The augmented images in some case has a different ratio/size compared to original image. That is a problem when the image shall be elaborate by anomaly that expect to have all images in same resolution
• The augmentation can’t be tuned with parameters (i.e. change max rotation to apply, max shift, etc), that make difficult to apply to specific uses case
• The augmentation often create images highly distorted, introducing big black areas or pixels distorsion. That make difficult to train the anomaly with augmented images