Deliver precise visual search for highly specific queries. hyper3-clip keeps hierarchy in the embedding space, helping rank the exact matches standard models miss.
See how hyper3-clip and OpenAI-CLIP rank exact matches when hierarchy and fine-grained details matter.
“grey velvet tufted sofa, brass legs, mid-century”
Rivet Freemont Sofa
Rank #1 • Sofa
Rivet Uptown Sofa
Rank #2 • Sofa
Rivet Freemont Sofa
Rank #3 • Sofa
Rivet Freemont Sofa
Rank #4 • Sofa
Rivet Uptown Sofa
Rank #5 • Sofa
Rivet Uptown Sofa
Rank #1 • Sofa
Rivet Alden Sofa
Rank #2 • Sofa
Rivet Uptown Sofa
Rank #3 • Sofa
Rivet Bradford Sofa
Rank #4 • Sofa
Rivet Accent Chair
Rank #5 • Chair
hyper3-CLIP retrieves the exact velvet variant at #1. OpenAI-CLIP buries it at #20 and bleeds into dining chairs.
Public benchmarks against OpenAI-CLIP. hyper3-clip’s largest gains are on same-item, variant, and exact-match retrieval — the cases where catalog search usually breaks. Broad category retrieval is closer, with one coverage metric still favoring CLIP.
Swipe to compare →
| Industry / Dataset | Benchmark | hyper3-CLIP | OpenAI-CLIP | Readout |
|---|---|---|---|---|
| Ecommerce Catalog Retrieval – Amazon Berkeley Objects | ||||
| Retail catalogs500 product images, 20 product types | Product-type mAP | 0.582 | 0.552 | +3.05 pts |
| Retail catalogs50 parsed catalog departments | Department mAP | 0.264 | 0.212 | +5.20 pts |
| Retail catalogsParent category retrieves diverse children | Child coverage@50 | 0.780 | 0.655 | +12.50 pts |
| Fashion Product Search – DeepFashion In-Shop | ||||
| Apparel retailSame-item product image retrieval | mAP | 0.407 | 0.240 | +16.7 pts |
| Apparel retailSame-item first-result recovery | Recall@1 | 0.595 | 0.375 | +22.0 pts |
| Apparel retailSpecific typed product search | Hit@10 | 0.572 | 0.550 | +2.2 pts |
| General Visual Hierarchy – COCO Objects | ||||
| Object search5,000 COCO val images, 80 categories | Category mAP | 0.554 | 0.532 | +2.22 pts |
| Object search12 object supercategories | Supercategory mAP | 0.536 | 0.516 | +2.08 pts |
| Object searchCoverage of child types under broad labels | Child coverage@100 | 0.887 | 0.951 | CLIP +6.40 pts |
hyper3-clip is an image-text embedding model trained on a hyperbolic manifold, giving structured visual data more room than it gets in a flat embedding space.
HyperView is an agent-native workbench for inspecting embedding spaces, curating datasets, and understanding why retrieval results fail.
Send a small sample of your images and queries. We run hyper3-clip against your current baseline and return a short report within 48 hours: metrics, ranked examples, and whether a pilot is worth it. No discovery call required.
Why mainstream ML's reliance on Euclidean manifolds is a mistake, and how hyperbolic spaces can efficiently encode hierarchical data with fewer dimensions.
read post →
View model