Oindrila Saha

Facing an energetic audience at the Los Angeles Convention Center in late October, Oindrila Saha took the stage at Adobe MAX Sneaks 2025 to present her research on AI-assisted scene generation.  

Saha, a doctoral student at the Manning College of Information and Computer Sciences (CICS) who also worked as a research scientist intern at Adobe, unveiled Project Scene It, an AI tool that gives creators greater control over how assets are placed within generated scenes. 

“Presenting Project Scene It was an invaluable experience—it helped tremendously with visibility for our work,” Saha said. “Our rehearsals, the opportunity to connect with talented professionals, and the exhilarating experience of presenting in front of a large, enthusiastic audience were incredible.”

While the idea for Project Scene It emerged during Saha’s internship, the machine learning model behind the tool is directly connected to the PhD research she’s completing at CICS.  

Her dissertation, “Fine-Grained Reasoning with Limited Supervision,” explores additional ways to improve visual recognition with less human supervision by harnessing generative models, natural language, and structured visual representations. 

“My coursework at CICS offered foundational perspectives that supported the learning I undertook for this project,” Saha said. 

The Emergence of Project Scene It

Raised in Noida, India, Saha first developed a strong interest in computing research while pursuing her bachelor’s degree in electronics engineering at the Indian Institute of Technology Kharagpur.  

After earning her undergraduate degree, she spent two years as a research fellow at Microsoft Research, where she realized she wanted to pursue a PhD and conduct further research in machine learning (ML).  

“I was drawn to the University of Massachusetts Amherst for the active ML research landscape and also because I was interested in the work my current PhD advisor, Professor Subhransu Maji, was exploring,” she said. 

Saha applied to CICS and began her doctoral program in 2021. In April 2025, she began interning at Adobe, where she and her colleagues identified a lack of methods for inserting reference objects into images while controlling their position and structure—a problem that led to the development of Project Scene It. 

“Coming up with a solution required detailed ideation and experimentation,” Saha said. “I came up with a strategy for synthetic data generation and model adaptation to develop the underlying method, and my mentors and collaborators provided crucial guidance towards the development of the project.” 

Project Scene It enables users to precisely control the placement and orientation of multiple objects in image generation, using a single reference image per object. By leveraging a text-to-image generation model using multiple modalities, creators can manipulate the generated image freely. 

Imagine a shoe company wants to showcase its newest sneakers on a mountaintop at sunset. Project Scene It takes a 2D image of the sneakers and places an accurate 3D model into the desired scene—all while giving the creators the ability to manipulate the size of the shoes, add in other assets, and change the background however they see fit. 

“Project Scene It can be very useful for designers who need to showcase their specific assets in varied contexts,” Saha said. “This direction of work can reduce the significant time normally spent arranging and photographing custom setups.” 

Overall, Saha found her Adobe internship experience was “very enriching.” Not only did she get timely feedback and advice from her collaborators, but her work landed her the opportunity to speak on stage at the Adobe MAX Sneaks event—an honor that led to a job offer. 

“Adobe Research is working on a range of very exciting and impactful problems in computer vision,” she said. “Researchers at Adobe are exceptional and very supportive of students' interests. This experience has helped me reevaluate how to approach research problems with user perspectives in mind.” 

The Power of Hands-On Learning

While CICS has provided Saha with essential foundational knowledge, she said that practical experience was crucial for developing her skills. 

“Hands-on experience in the direction that one is interested in pursuing after graduation is very important to understand what to expect and how to prepare for life after a PhD,” she said. “I have gained more insight about user-centric problems where our research can be applied and how to approach them.” 

She encourages fellow doctoral students—and even those who are earlier in their academic careers—to build networks with alumni and stay informed about opportunities beyond academia.  

“While looking for opportunities, I encourage my fellow PhD students to stay informed about our strong UMass alumni community, many of whom are doing outstanding work in leading organizations and research areas,” she said.  

Reflecting on her time at CICS, Saha highlighted the importance of mentorship and collaboration in shaping her academic and career path. 

“I have been immensely lucky to be advised by brilliant professors at the computer vision lab and collaborate with the bright PhD students in our lab,” she said.