InternVL: Scaling Up Vision Foundation Models

摘要

InternVL scales up vision foundation models to 6 billion parameters. The model achieves state-of-the-art on diverse visual benchmarks including recognition, detection, and segmentation.