Niladri Shekhar Dutt

Tell us a bit about you   

I am a PhD student in Computer Science at UCL, conducting research at the intersection of machine learning, computer vision, and computer graphics.   

Originally from Kolkata, India, my early fascination with creating computer games as a kid, and programming during high school, sparked my interest in the field of computer graphics and machine learning.   

Prior to beginning my PhD, I completed my MSc in Computer Graphics, Vision and Imaging here at UCL.  

What problems are you trying to solve with your research?   

My research aims to bridge the gap in domain knowledge for large language models (LLMs) when applied to image and 3D tasks. Existing LLMs such as ChatGPT lack complex visual understanding and manipulation.  

My recent work, MonetGPT, addresses this by exploring how we can train LLMs using specially designed visual puzzles. By learning to solve these puzzles with reasoning, LLMs can acquire expertise in areas like image enhancement/retouching, enabling them to edit images much like an expert using tools such as Lightroom or Photoshop.  

A significant benefit of this approach is its inherent explainability. By integrating LLMs with a library of editing operations, not only does a user receive the final, enhanced/retouched image, but they also get detailed explanations for each step taken.  

This transparency builds trust in AI and significantly enhances learnability for novice users. I am excited to present this work at SIGGRAPH, a leading annual conference in computer graphics, being held in Vancouver, Canada this August.   

Looking ahead, I aim to explore the integration of 3D domain expertise into LLMs, leveraging their inherent reasoning capacities to enhance both the generation of novel 3D content as well as understanding of 3D environments.  

You are also a part of the ELLIS (European Lab for Learning & Intelligent Systems) PhD programme. What are the benefits of this?

The ELLIS PhD programme acts like a meta programme, and it provides a unique structure, pairing students with a co-advisor from either industry or a university in a different European country.  

A key requirement is spending at least six months working with this advisor apart from regular collaboration. My co-advisor is Dr. Duygu Ceylan at Adobe Research. Having her as a collaborator is incredibly valuable, and I am interning with her this summer.  

The ELLIS network itself is a major asset, comprising many leading researchers and offering numerous events and workshops throughout the year. The programme also provides dedicated funding for travel and mobility, which is quite helpful.  

Attending last year's ELLIS Doctoral Symposium, for instance, allowed me to connect with students from across Europe, exchange ideas, build friendships, and gain new perspectives. ELLIS Unit London is hosted at UCL.  

What’s an average day like for a PhD student at UCL Computer Science?  

I enjoy being in the lab, where I can discuss my research with my advisor and fellow lab mates. Typically, my mornings involve meetings, which become more frequent when I'm actively collaborating on projects.  

I usually start my day by reading new research papers in my field. Following that, I focus on making progress towards my weekly research goals, often by designing and running new experiments or analysing my previous experiments and solving bugs.   

Where do you see yourself after the PhD?  

I am currently weighing my options between pursuing a career in academia or moving into industry research.  

If I choose the academic path, I plan to continue with a postdoctoral position with the eventual goal of becoming a professor and leading my own research lab. Alternatively, I might move into industry research as a research scientist and continue publishing my work.  

What advice do you have for others that might be considering a PhD at UCL Computer Science?  

Choosing the right research lab is arguably the most critical decision when considering a PhD. Several factors contribute to this, including finding a good fit with your potential advisor's mentoring style and your prospective lab mates, ensuring alignment with your own research interests and goals, and establishing a healthy understanding of work-life balance and research expectations.  

It is essential to have open and honest discussions with your potential advisor about these aspects before making your decision. UCL Computer Science has several world leading research labs, which makes it a great place to pursue a PhD.   

Which part of MSc Computer Graphics, Vision and Imaging best prepared you for what you’re doing now?  

I feel the entire MSc programme in Computer Graphics, Vision and Imaging provided an excellent foundation for my current research in computer vision and graphics. The strategic combination of modules made it easy to connect concepts across different areas.  

I particularly recall how studying neural rendering (like NeRFs) brought together elements from all four of my core modules from the first term, leading to a profound understanding of the topic.  

What was the most rewarding part of your MSc?   

The MSc thesis project was particularly rewarding as I had the opportunity to join the Smart Geometry Processing Group led by Professor Mitra. This is the same group where I am now pursuing my PhD.  

The reading group sessions were incredibly insightful and helped me navigate the broader research landscape. I thoroughly enjoyed working on my thesis and later presented it at CVPR '24 held in Seattle, USA.  

What surprised you most about the teaching style at UCL Computer Science?  

I valued the strong focus on depth over breadth. Rather than rushing through a wide syllabus, the course emphasised a deeper understanding of core principles, often going over difficult concepts multiple times. It gave me the foundation I needed to explore new areas on my own, a skill I find essential for research.   

The information on this page reflects the student's status at the time of publication (June 2025).