Brown CS PhD Student Rahul Sajnani And Co-Authors Receive A WACV Best Student Paper Award
- Posted by Jesse Polhemus
- on March 25, 2025

Brown CS doctoral student Rahul Sajnani has just been honored with the Best Student Paper Award for his research (“GeoDiffuser: Geometry-Based Image Editing with Diffusion Models”) at the 2025 Institute of Electrical and Electronics Engineers (IEEE)/Computer Vision Foundation (CVF) Winter Conference on Applications of Computer Vision (WACV). WACV is often considered one of the most prominent international conferences devoted to computer vision. Coauthored with Jeroen Vanbaar, Jie Min, and Kapil Katyal of Amazon Robotics, as well as Brown CS faculty member Srinath Sridhar, the paper introduces a novel approach to image editing by unifying common 2D and 3D object editing capabilities into a single method.
GeoDiffuser presents a zero-shot optimization-based method that views image editing operations as geometric transformations. By incorporating these transformations directly into the attention layers of diffusion models, the method performs precise edits without the need for additional training or model fine-tuning. This approach allows for common edits like object translation, 3D rotation, and removal, addressing limitations in existing methods that are often bespoke, imprecise, or limited to 2D edits as shown in the given image.
“One aspect of this work that I really like,” says Rahul, “is that it only involves carefully manipulating features within transformer blocks based on geometry existing in today’s generative models and test time optimization to produce edits that remove objects, move them around, and rotate them. Transformer blocks are the building blocks of today’s generative models that attend to image and language tokens for better generation. This work would not have been possible without my great team from both Amazon Robotics and Brown University as well as great works from the community that shape our understanding of generative models.”
“Generative AI has taken the world by storm,” Srinath notes. “Although current AI image generators are very capable, controlling or adjusting the spatial layout of specific objects in an image is difficult. Rahul’s work solves this challenging problem.”
Please find the project page with all the edits and link to the GitHub code here.
A list of all award winners from the conference is available here.
For more information, click the link that follows to contact Brown CS Communications Manager Jesse C. Polhemus.