Presentations · 2026.02.25
Clustering-Based Dataset Curation
Clustering-Based Dataset Curation
Overview
Research poster on reducing redundant driving images and overrepresentation of common scenes using CLIP embeddings, UMAP and two-stage HDBSCAN clustering. Compares cluster-based sampling strategies in a front/rear camera frame-matching experiment.
Version: nxtAIM Open Project Day original research poster; one page