finetheAi R&D Center
finetheAi Product & Solution R&D CenterCompany website finetheai.com ↗
From seeing
to deciding.
Data collection, training, line-side inference, operations and robotics. Installers and technical material for the vision AI software built at the finetheAi lab.
yolo_alignRotates the part to 12 o’clock and centers it, then YOLO finds the tab and any defect. Used on the EB line since v2.2.
One pipeline, five stages
Every product follows the order in which vision AI reaches the line. The output of one stage is the input of the next.
Data collection & preprocessing
Model design & training
Evaluation & line inference
Recent releases
Builds dropped into the internal release folder are checked and published automatically.
| Released | Product | Stage | Version | Size | downloads | |
|---|---|---|---|---|---|---|
| 2026-10-16 | fineAi RoboVision | Physical AI | 2.3.0 · 261016b | 3.10 GB | 1 | Download → |
| 2026-10-16 | fineAi RoboVision | Physical AI | 2.3.0 · 261016 | 3.05 GB | 1 | Download → |
| 2026-10-06 | fineAi Viewer & Inspector | Data collection & preprocessing | 2.3.4 · 261006 | 202 MB | 0 | Download → |
| 2026-09-16 | fineAi Multi-Inspector (EB) | Evaluation & line inference | 2.3 · 260911 | 2.49 GB | 0 | Download → |
| 2026-09-07 | CUBE FARM | Physical AI | 1.0 | 2.33 GB | 0 | Download → |
| 2026-08-31 | fineAi Navigator | Operations & monitoring | 1.2.0 | 39 MB | 0 | Download → |
| 2026-08-22 | fineAi Viewer & Inspector | Data collection & preprocessing | 3.1.0 | 2.27 GB | 0 | Download → |
| 2026-08-22 | fineAi Multi-Inspector (CT) | Evaluation & line inference | 3.2.1 · 260821 | 2.39 GB | 0 | Download → |
Notes from the lab
What the lab learned on the line — algorithms, performance work and installation know-how.
The bottleneck was the PNG encoder — a 16× faster augmentation
Over 90% of augmentation time went into PIL's PNG encoder. One compression level and parallel saving took each image from 4.3 s to 0.26 s.
Aligning parts that are not round — moments and principal axes
Training is more stable when every part faces the same way, round or not. How first-, second- and third-order moments give position, orientation and front/back.
Straighten first, then judge — Multi-Inspector's yolo_align mode
A new mode that runs YOLO on images that are only rotated and centered, without Canny preprocessing — reproducing training-time alignment on the line.
An AI that has read every document answers first
Answers cite release notes, manuals and installation guides. If it cannot solve your problem, the thread goes straight to an engineer.