Intra Video Positive Pair

Intra-video positive pairs (IVPP) in self-supervised learning leverage the temporal relationships within videos to create training data for various tasks. Research focuses on defining effective similarity metrics between video frames (e.g., proximity in time) and incorporating these pairs into contrastive or non-contrastive learning frameworks, often employing weighting schemes to emphasize closer temporal relationships. This approach addresses challenges in data scarcity and computational cost associated with traditional negative sampling methods, improving performance in applications like ultrasound image classification and action sequence verification. The resulting improved representations benefit downstream tasks by enhancing intra-class compactness and leading to more robust and accurate models.