CyberVision is a framework for solving Manual Inspection problems that have a complex rules process that can found in the heavy machinery, automotive, and robotics spaces. CyberVision is built with two modes of deployment in mind. The first being a pure AWS pipeline consisting of an AWS IoT GreenGrass deployment, an Amazon Lookout for Vision model, and a SageMaker Ground Truth annotation pipeline and the second being a software library complete with model compression and tuning pipelines capable of being run on both CPU and GPU devices.

Defect Detection

CyberVision utilizes both the Lookout for Vision Segmentation and Classification system along with the ability to utilize State-of-the-Art segmentation and detection models that can be compressed and deployed to a given deployment target.

Model Compression

Weight compression and speed increase

Backend, Models, and Support

Training

BYOM

python3 train_segmentation.py

Lookout for Vision

...
...

l4v.fit(
    output_bucket=output_bucket,
    model_prefix="mymodelprefix_",
    wait=True)
...

Architecture