Dashboard
Readiness
Guidance before training; an admin may train earlier. Promotion always needs the benchmark gate: at least 3 test documents and 100 reference points on real drawings.
Active models
Label queue
Pages with the most uncertain pins first. Review opens the viewer on that pin; mark a page fully reviewed once every receptacle on it is pinned and no pin is left unreviewed.
| Drawing | Page | Unreviewed | Most uncertain | Status |
|---|
Nothing left to review. Scan more pages in the viewer.
Most uncertain pins
| Drawing | Page | Score | Distance from threshold |
|---|
Datasets
A dataset is a snapshot of the reviewed pins and fully reviewed pages, split by drawing into train, val and test.
| Dataset | Created | Verifier crops (+/-) | Detector pages (points) | Test split |
|---|
No datasets yet.
Training runs
The server shuts down at 7 PM Eastern on weekdays and over the weekend. A run still going then is interrupted and has to be started again, so start long runs in the morning.
| Job | Kind | Status | Progress | Started by | Created |
|---|
No training jobs yet.
Log
Models
No models yet. Train one from a dataset.