View model
ECCV BEW 2026
Hyper3-CLIP: hierarchy-conditioned hyperbolic vision-language training
Query-conditioned visual pooling and hierarchical training in hyperbolic space.
Your data is hierarchical.Your model should be too.
Deliver precise visual search for highly specific queries. hyper3-clip keeps hierarchy in the embedding space, helping rank the exact matches standard models miss.
Where exact matches get buried.
See how hyper3-clip and OpenAI-CLIP rank exact matches when hierarchy and fine-grained details matter.
Browse all HyperView SpacesMore demos
Find the right product by material, color, and style.
“a product photo of grey velvet sofa”
- metal
- brass finish
- wood
- velvet upholstery
- tufted
Exact query
a product photo of grey velvet sofa: Frederick Mid-Century Modern Tufted Velvet Sofa Couch 77.5 W Grey, with metal, brass finish, wood, velvet upholstery, tufted
Target product
Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5 W, Grey
hyper3-CLIP
Exact product #1- #1
Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5"W, Grey
sofa · Grey Velvet
Exact product 
#2Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Charcoal & Brass
sofa · Charcoal
#3Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5"W, Forest Green
sofa · Forest Green Velvet
#4Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5"W, Navy Blue
sofa · Navy Blue Velvet
#5Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Dove Grey & Brass
sofa · Dove Grey
OpenAI CLIP
Exact product #5
#1Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Hunter Green & Brass
sofa · Hunter Green
- #2
Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Charcoal & Brass
sofa · Charcoal
#3Rivet Uptown Mid-Century Velvet Tufted Customizable Daybed Sofa, 78"W, Shadow & Silver
sofa · Shadow
#4Rivet Alonzo Contemporary Leather Sofa Couch, 80"W, Grey
sofa · Grey Leather
- #5
Rivet Frederick Mid-Century Modern Tufted Velvet Sofa Couch, 77.5"W, Grey
sofa · Grey Velvet
Exact product
See how your own catalog compares.
Get a 48-hour evaluationWhy Geometry Matters
Standard models cram your data into a flat (Euclidean) space. This creates a fundamental mathematical mismatch:
- Hierarchical data grows exponentially.
- Euclidean space grows only polynomially.
- The Mismatch: Flat models force exponential hierarchies into polynomial dimensions, crowding siblings together and causing retrieval failures like category bleed.
Euclidean Space
Hyperbolic Space
Model results
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 Matching & Search – DeepFashion In-Shop | ||||
| Apparel retail710 photo queries, 741 catalog photos | mAP | 0.635 | 0.352 | +28.3 pts |
| Apparel retailSame product across views; query photo excluded | Recall@1 | 0.759 | 0.456 | +30.3 pts |
| Apparel retail180 text queries, separate 1,120-photo pool | 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 |
Fashion photo-matching results use the same catalog as the Fashion Matching Space. Text search uses a separate evaluation pool.
hyper3-CLIP
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
HyperView is an agent-native workbench for inspecting embedding spaces, curating datasets, and understanding why retrieval results fail.
Get a 48-hour retrieval eval.
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.
01Copy the prompt into your assistant
02Answer three questions, send the email
03Get an eval report back within 48 hours
The Geometry Mistake Behind Modern Embedding Models
Why mainstream ML's reliance on Euclidean manifolds is a mistake, and how hyperbolic spaces can efficiently encode hierarchical data with fewer dimensions.
