An Annotation Robot

This is my working note, will be updated dynamically…

Simple idea:

explanation: focus on the non-shadow part, release human from the tedious labelling work.

We need a micro-fixed-task model (non-overlapped model, overlapped model…), not a big-generic model:

Learning during the human’s labeling, not afterword:

Finding the similarities to label, not limited to Euclid, cosin or dot’s, also entropic, and micro-model, which executed by a 200-lines-AI-os, this AI-os even can run most tensorflow, or PyTorch model:

Architecture (How it works with pieces of code blocks):