V-JEPA 2 Data Pipeline

What problem does it solve?

This Skill helps ML engineers build and validate a scalable V-JEPA 2 video data pipeline, coordinating video decoding, clip sampling, augmentation, multi-source data mixing, and deterministic data loading.

Core Features & Use Cases

• Multi-source VideoDataset support with fps, duration, and frame_step clip modes
• YAML-config driven augmentation and data loading via DataConfig and AugConfig
• Deterministic, reproducible DataLoader pipelines with per-source weighting and distributed sampling
• Ready-made templates for data management, transforms, and evaluation, plus synthetic data fallbacks for testing

Quick Start

Create a DataConfig with your data_paths and an AugConfig for augmentation, then instantiate DataManager to build the train loader.