Skip to main content
Back to help

Troubleshooting

5 min read

Fixing low symbol recognition accuracy

If recognition is below 70% on your trade, the active-learning loop hasn't seen enough of your drawings yet.

If the recognizer is missing more than 30% of symbols on your trade after the first few projects, the per-org model hasn't accumulated enough training data.

Diagnose first

  1. Go to Symbols → Training dashboard.
  2. Check the per-trade recognition trend. Is it climbing or flat?
  3. Check 'Annotations contributed'. Below 100 in your trade means the training set is too small.

Accelerate

  • Walk through three legacy projects and confirm/reject every queue item — fastest way to add 100+ annotations
  • Use the bulk-import tool: upload a folder of legends + reference projects, the recognizer extracts symbol crops automatically
  • Mark your most-used symbols as 'priority' in the Symbol Library — the trainer weights them more heavily
  • Wait one retrain cycle — Saturdays at 02:00 UTC

Still stuck?

Contact support — one business hour reply.

Contact supportBrowse all help articles
Fixing low symbol recognition accuracy — OmniTakeoff Help