Board Count AI v2 — Faster, More Accurate
Major accuracy and speed improvements to our board counting AI model, now handling stacked and overlapping boards with 97%+ accuracy.
We've shipped a significant upgrade to our board counting AI model.
What's New
- 97%+ accuracy on stacked and overlapping boards — up from 93% in v1
- 2x faster inference on edge hardware, reducing latency from ~200ms to ~90ms per frame
- Improved low-light performance for operations running dimly lit trim saw areas
- Auto-calibration — the model now self-adjusts to camera angle changes without manual reconfiguration
Why It Matters
Board counts have historically been the hardest metric to track accurately. Boards overlap, stack irregularly, and move fast on conveyors. This update closes the gap between board counting and our already-strong pallet counting accuracy.
Rollout
This update is live on all edge devices. No action needed — models are updated automatically during the next sync window.