Dragonfly: A large vision-language model with multi-resolution zoom

Dragonfly, a new vision-language model, uses a breakthrough multi-resolution zoom-and-select architecture to enhance visual understanding. It achieves improved performance on commonsense visual QA and image captioning benchmarks.

• Dragonfly unveiled by Together AI.
• New model uses multi-resolution zoom.
• Achieves fine-grained visual understanding.
• Open-source models with competitive results.
• Developed with Stanford Medicine partnership.