Feature Extraction
Image Embedding
Generate 768-dimensional SigLIP embeddings from images for visual similarity search
Why do anything?
Images need vector representations for visual similarity search. Without embeddings, you can't find visually similar images.
Why now?
Visual search is expected in e-commerce, media, and content moderation. Users search by example images.
Why this feature?
SigLIP model produces high-quality 768D visual embeddings. Supports image preprocessing and batch processing.
How It Works
Image extractor uses SigLIP model for visual embeddings optimized for image retrieval and similarity.
1
Preprocessing
Resize, normalize, format conversion
2
Embedding
Generate 768D SigLIP embeddings
3
Storage
Store in Qdrant with vector index
Why This Approach
SigLIP provides excellent visual-semantic alignment. 768D provides good quality/cost tradeoff.
Where This Is Used
Integration
client.collections.create(feature_extractor={"feature_extractor_name": "image_extractor", "version": "v1"})