Model under test
No completed training runs yet — train a model first.
Want to try the model on your own photos or videos? Use the Lab page.
Metrics on a held-out split
—
mAP@[.5:.95]
—
mAP@50
—
mAP@75
—
mAR@100
| class | instances | AP@50 | AP@[.5:.95] |
|---|
Error analysis
Run an evaluation on a split first — the analysis is computed from its saved detections.
0.50
—
correct
—
missed
—
false positives
—
wrong class
Confusion matrix — rows: ground truth, columns: predicted
Per class
| class | recall | precision | missed | false | confused |
|---|
Most confused pairs
Worst images
ground truth (matched)
correct
false positive
missed
wrong class
No errors at this threshold — every ground-truth box is found and nothing is hallucinated.