Find the Leak, Fix the Split: Cluster-Based Method to Prevent Leakage in Video-Derived Datasets

TL;DR

This paper introduces a cluster-based frame selection strategy to prevent information leakage in video-derived datasets, ensuring more representative and balanced dataset partitions for improved model evaluation.

Contribution

It presents a novel clustering approach to improve dataset splitting, reducing leakage and enhancing the reliability of video-based machine learning evaluations.

Findings

Reduces information leakage in video datasets

Creates more balanced and representative dataset splits

Improves reliability of model evaluation on video data

Abstract

We propose a cluster-based frame selection strategy to mitigate information leakage in video-derived frames datasets. By grouping visually similar frames before splitting into training, validation, and test sets, the method produces more representative, balanced, and reliable dataset partitions.