Making Deep Learning Workflows, Relational

Continuing our theme of built-in UDFs, we preview Segment Anything Model UDF as a foray into Rust’s ML capabilities with HuggingFace’s candle Library. We show how to integrate the SAM model into a relational workflow, enabling portability and complex data processing.

In this post, we build on top of Rust-based ML ecosystem, showing the capabilities of candle as performant integration with Python. We turn a PyTorch+pandas workflow with the Segment Anything Model (SAM) into relational operations; this enables portability and performance. We show 4x speed-ups over the external Pytorch/pandas based workflow. Finally, we explore pa.binary type for representing images.

Unstructured data, including text documents, videos, and images, constitutes an overwhelming 90% of the world’s information. Despite its prevalence, 95% of organizations need help to effectively manage and extract value from this data. This challenge is particularly acute when it comes to image data.

The impact of successfully integrating image analysis with traditional data processing could be transformative across various industries:

Leave a Comment