Vaibhav Agrawal supervised by Dr. Ravi Kiran Sarvadevabhatla  received his Master of Science – Dual Degree  in Computer Science and Engineering (CSD). Here’s a summary of his research work on  Multi-Object 3D Control in Text-to-Image Generation:

This thesis focuses on improving fine-grained control in text-to-image generation, particularly for complex scenes containing multiple objects. While existing generative models can create realistic images, they often lack the ability to precisely control the arrangement, orientation, and interactions of individual objects. To address this, the thesis introduces Compass Control (CVPR 2025), the first framework for multi-object 3D orientation control, enabling users to specify the orientation of each object independently through specialized prompt tokens. Building on this work, the thesis presents SeeThrough3D (CVPR 2026), a novel approach for 3D layout-conditioned image generation that explicitly models inter-object occlusions using an occlusion-aware scene representation. Together, these contributions enable more realistic and controllable scene generation, advancing the state of the art in 3D-aware image synthesis and providing new tools for creating complex, multi-object visual scenes.

May 2026