niladridutt.bsky.social
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/
40 posts
1,890 followers
738 following
Regular Contributor
Active Commenter
comment in response to
post
🧵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
comment in response to
post
🧵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).
comment in response to
post
🧵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.
comment in response to
post
🧵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
comment in response to
post
🧵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.
comment in response to
post
🧵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.
comment in response to
post
🧵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.
comment in response to
post
🧵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.
comment in response to
post
🧵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
comment in response to
post
🧵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
comment in response to
post
🧵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).
comment in response to
post
🧵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.
comment in response to
post
🧵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
comment in response to
post
🧵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.
comment in response to
post
🧵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.
comment in response to
post
🧵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.
comment in response to
post
🧵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.
comment in response to
post
🧵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
comment in response to
post
Amazon came pretty late to India and we already had some homegrown companies like Flipkart which now competes with Amazon and is valued at $40B.
I think big tech's early access killed homegrown companies. China and to an extent South Korea (Naver) has some great tech companies because of barriers
comment in response to
post
Who will tell the silicon valley tech bros that it wasn't them alone
comment in response to
post
Haha exactly what I did today!
comment in response to
post
Have a great time in Seattle!
comment in response to
post
Added you!
comment in response to
post
Added you!
comment in response to
post
Added you!
comment in response to
post
👋
comment in response to
post
Hey Orion
Thanks for the great least. Currently working on RAG, could I be added as well?
comment in response to
post
While the following feed doesn't depend on likes since it follows a chronological order, I think the discover feed uses likes to tune your experience + recommend popular content
comment in response to
post
I think keeping likes anonymous allows people to freely like whatever content they want. Not liking reduces post engagement. It also allows one to freely like content without it popping on someone else's feed (I guess this is less of a problem here since only retweets are shown to your network)
comment in response to
post
As an author Twitter allows you to see who liked your posts/replies but it's anonymous to others.
comment in response to
post
Hi Kosta,
Could you add me too?
comment in response to
post
Exactly, not getting penalized for external links is a big win!
comment in response to
post
Sure, added you :)
comment in response to
post
Thanks for the list! I've created one for inverse graphics and 3D vision-- bsky.app/starter-pack...