World to Code: Multi-modal Data Generation via Self-Instructed Compositional Captioning and Filtering

VLM
VGen
Re-assign community