The Geometry of Image Embeddings, Hands-on Coding Workshop, Part I

```
Join us for a deep dive into vector space!

In modern AI, transforming images, text, and any other modality into dense vector representations (embeddings) is one of the most powerful tools in a practitioner’s arsenal. But how do we actually leverage the geometry of these high-dimensional spaces to solve real-world problems?

Sponsored by KINETO.AI, this is a hands-on coding workshop designed for ML engineers, data scientists, and computer vision practitioners. We will move beyond theory to code practical solutions using PyTorch.

Special Feature: Introducing HyperView We are excited to present HyperView, a new open-source dataset curation tool. During the session, we will use HyperView to visualize and interact with image embeddings across different geometric manifolds, including:

• Standard Euclidean projections.
• Hyperspheres (for cosine similarity optimization).
• Hyperbolic spaces (for hierarchical data representation).

What we will explore: In this session, we will code through three specific use cases:

• Similarity Search: Learn how to map images into a vector space where geometric distance equals semantic similarity.
• Dataset Curation with HyperView: Use the tool to visualize data distribution, detect outliers, and identify duplicates using advanced geometric projections.
• Re-identification (ReID): Implement pipelines to recognize and match the same object or person across different images.
  ```