COCO 2017: Pre-Computed Image Embeddings Using SigLIP 2

Pre-computed image embeddings for the COCO 2017 dataset, generated with Google's SigLIP 2 vision model. The embeddings are L2-normalized 1152-dimensional vectors derived from images resized to 384x384 pixels. This dataset was created by jrmiller and last updated on Hugging Face in April 2026.

Use Cases

• Image retrieval and similarity search based on the 1152-dimensional embedding vectors.
• Training or fine-tuning downstream models using pre-extracted features from the SigLIP 2 model.
• Benchmarking embedding models and techniques on the standard COCO 2017 dataset.
• Conducting multimodal analysis by linking these image embeddings with corresponding COCO captions or annotations.

Strengths

• Embeddings are L2-normalized, which can simplify distance calculations for tasks like similarity search.
• Generated using a specific, documented model (google/siglip2-so400m-patch14-384).
• Based on the widely recognized and annotated COCO 2017 source dataset.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.

Provenance

Source

Hugging Face user jrmiller.

Collection Method

Pre-computed using Google's SigLIP 2 model on images from the COCO 2017 dataset.

Time Range

Based on the COCO 2017 source dataset.

Freshness

Last updated 2026-04-12 22:00:53; freshness should be verified.

Geography

null