Research Intern @adobe.com | PhD @ucl.ac.uk | @ellis.eu | ex-Nvidia, Berkeley | Interested in generative modelling in vision and graphics + reasoning (LLMs)
https://niladridutt.com/
Niladri Shekhar Dutt
🧵10/10 Lastly, huge thanks to my co-advisors Niloy and Duygu!
For more details check out our paper below-
🌐 Project Website: monetgpt.github.io
📄 Arxiv: arxiv.org/abs/2505.06176
🧵3/10 Our key recipe: MLLMs struggle to predict edit values directly. We solve this by generating rich textual reasoning for each puzzle ✍️. We then fine-tune MonetGPT on this data, creating a 'reasoning pathway' that enables it to regress final adjustment values.
🧵7/10 Our puzzle-based training with a 'reasoning as a pathway' approach allows MonetGPT to generate detailed justifications for each edit, delivering truly explainable image retouching
🧵2/10 MLLMs lack the visual understanding to plan edits. 🧠 So, we use expert photos as our ground truth and work backward, procedurally creating puzzles by assuming any change to an expert edit makes it less optimal
🧵1/10 Excited to share our #SIGGRAPH paper "MonetGPT: Solving Puzzles Enhances MLLMs' Image Retouching Skills" 🌟
We explore how to make MLLMs operation-aware by solving visual puzzles and propose a procedural framework for image retouching
#MLLM
🧵9/10 We quantitaively evaluate on the Adobe5k dataset as well as conduct user studies by expert and novice users. Our evaluations show that MonetGPT outperforms open-source alternatives and performs comparably to Google Photos AutoEnhance (closed-source).
🧵8/10 Photo editing is subjective 🎨. Our framework adapts to user preference by guidance from natural language tags like ‘vibrant’ or ‘retro vibe’ to produce personalized and stylistically distinct retouching plans from the same input image.
🧵4/10 🧩 Puzzle A builds an understanding of individual operations. The MLLM learns to map visual changes in before/after images to a specific tool and its precise parameter value, effectively learning the semantics of our procedural library.
🧵6/10 🧩 Puzzle C builds planning capabilities. The model learns to generate a complete, multi-step retouching plan to enhance a photo, structuring its reasoning as a sequence of discrete issues and solutions for clarity and control.
🧵5/10 🧩 Puzzle B imparts aesthetic judgement. By ranking professionally edited photos against altered versions, the MLLM learns to recognize the visual characteristics of an optimally adjusted image for any given operation, building an internal aesthetic model.
