{readout}
-{readout}
-{mode === "photo" ? "Use one product photo to find its other catalog views." : "Compare the same detailed request across both models."}
-“{active.query}”: null} - - - -
Correct means another view of the same product.
-| Correct product | Hyper3 | CLIP |
|---|---|---|
| Ranked first | {percentage(photoBenchmark.hyper3R1)} | {percentage(photoBenchmark.clipR1)} |
| Within top ten | {percentage(photoBenchmark.hyper3R10)} | {percentage(photoBenchmark.clipR10)} |
| Mean average precision | {Number(photoBenchmark.hyper3Map).toFixed(3)} | {Number(photoBenchmark.clipMap).toFixed(3)} |
Cases use the largest first-match rank gap in each direction, so the CLIP win is shown too. Photo matching and typed retrieval are separate evaluations; their scores are not interchangeable.
- > : <> -Correct means any view of the described product.
-| Correct product | Hyper3 | CLIP |
|---|---|---|
| Ranked first | {percentage(typed.hyper3Hit1)} | {percentage(typed.clipHit1)} |
| Within top ten | {percentage(typed.hyper3Hit10)} | {percentage(typed.clipHit10)} |
| Within top fifty | {percentage(typed.hyper3Hit50)} | {percentage(typed.clipHit50)} |
This is one illustrative Hyper3 win. Across the full 1,120-photo benchmark, the models are close and OpenAI CLIP is ahead at fifty results. The visible workspace map contains {props.workspaceSampleCount || 741} catalog photos.
- >} -{samplePage.error}
: null} - {error ?{error}
: null} -