GitHub · Serving and API
uu889/laya-opencv
Seeing a decision model used as the last step of a measurement pipeline rather than the first.
What it does
A local workbench that puts an image on one side and a verdict on the other: OpenCV does the segmenting, measuring and detecting, written rules settle anything numeric, and a Laya or Jev decision model answers the questions about the numbers — for fruit picking, weed removal, pest monitoring, steel pipe and textile inspection out of the box.
The repository's own description: “A one-click local workbench: OpenCV measures, hard rules settle the numbers, a Laya/Jev model gives the verdict.”
How it works
The measurements are turned into text and passed down a pipeline that ends in the model, so the model reads a sentence about an area or a ratio rather than pixels; disagreements between the rules and the model are flagged for human review instead of being averaged. The installer builds a virtualenv, picks a PyTorch build for the GPU it finds, and runs a local server that also speaks the systemone format.
The repository, by the numbers
Read from the GitHub API on 2026-10-09. Stars and the last push move daily — quote them with the date, the way we do.
What we checked
- the README's pipeline diagram and the five schemes it ships with, read 2026-10-09
- the 3.0 training page: 11 encoders, LoRA or full fine-tuning, CPU fallback, its own /v1/systemone server
- the install and start steps for Windows and Linux, and the pinned laya[serve] dependency
- the licence file and the GitHub metadata: 48 stars, 8 forks, last push 2026-10-08
What we did not check. We did not download the installer or run an inspection, so the screenshots, the training times and the claim that a small model trains on a 4 GB card are the author's and not something we reproduced.