CLIP Cross-Encoder

CLIP Cross-Encoder uses CLIP ViT-L/14 to compute pairwise similarity scores between images and text or between two images. Useful for reranking, visual similarity, and recommendation pipelines.

When to use:

• Image-to-image similarity reranking
• Image search reranking with text queries
• Use when CLIP embeddings are already in use in the pipeline

Input:

• Query Image (optional): Image to match against candidates
• Query Text (optional): Text query for image ranking
• Candidates (required): Candidate images to score

Output:

• Scores: Similarity or relevance scores per candidate
• Ranking: Indices sorted by relevance

Inference Settings

No inference-time settings. Scores are computed deterministically.