~/system-one-bench / benchmarks/refcoco
Vision track · Box questions · All benchmarks
Box the person or object a phrase describes, 100 people and 100 objects. Jev 1.14 is the only model on this benchmark so far.
| Rank | Model | Right (IoU ≥ 0.5) | 95% interval | Mean IoU | Calibration error | Failed | |
|---|---|---|---|---|---|---|---|
| 1 | Jev 1.14 TypeSafeLeader | 78.5 | 72.8–84.2 | 0.70 | 30.3% | 0 |
The bar is accuracy; the bracket on it is the 95% interval for 200 cases. Tied for 1st marks every model whose gap to the leader is within sampling noise (1.96 standard errors of the difference, treating the runs as independent). A failed call counts as wrong.
Box questions are scored apart and stay out of the Vision Index. Every model gets the same cases and the same request, under the benchmark rules.