Models Generic SigLIP2 SO400M

SigLIP2 SO400M

Vision encoder with Shape-Optimized 400M architecture and sigmoid contrastive loss.

Google Open Source Vision Only

HuggingFace arXiv

arXiv

Generic

Model Type

Overall Rank

45.68%

Avg. R@1

34.70%

Avg.
mAP@20

1152

Embedding
Size

384

Input
Size

Datasets

Dataset	Category	R@1	R@5	mAP
Stanford Online Products	E-commerce	80.28%	90.01%	60.79%
Products-10K	E-commerce	65.98%	84.25%	44.15%
TOPEX	Industrial	59.62%	78.23%	37.68%
Furniture v1	Retail	57.41%	76.82%	40.33%
DIY v1	Hardware/DIY	23.86%	46.61%	32.46%
Automotive v1	Automotive	22.02%	40.06%	25.32%
Clips-and-Connectors v1	Industrial	10.57%	23.12%	2.16%
Average		45.68%	62.73%	34.70%