Text-to-image generation for hard-to-acquire scenarios

Develop controllable text-to-image generation models for synthesizing data of rare, safety-critical, or otherwise underrepresented scenarios