Simone Foti

It's time to bring 3D meshes into modern machine learning properly! 🛸 Our work solves the non-differentiability of the Exp map on meshes, enabling gradients to flow directly through geodesics. It’s differentiable, GPU-fast, and fully parallelised. circle-group.github.io/research/DSG

Our #CVPR2026 work Differentiable Straightest Geodesics [circle-group.github.io/research/DSG/] provides exponential maps for 3D meshes allowing to trace geodesics or parallel-transport vectors. With it, we release DiGeo, a PyTorch differential geometry package: digeo.readthedocs.io. 👇

Huge thanks to the team for making this "straightest" vision a reality: @hippolytever (who developed this for his MSc thesis @ICComputing! 🙌), @Kausta42, Stefanos Zafeiriou, and @tolga_birdal. #CVPR #CVML #AI #Graphics #ML #PyTorch

🛠️digeo: for easy deployment, we make all of this available with our new PyTorch library for differential geometry on meshes. Docs & installation: digeo.readthedocs.io pip-install digeo and be part of the next generation of AI research on 3D meshes! 🚀

It's time to bring 3D meshes into modern machine learning properly! 🛸 Our work solves the non-differentiability of the Exp map on meshes, enabling gradients to flow directly through geodesics. It’s differentiable, GPU-fast, and fully parallelised. circle-group.github.io/research/DSG

Love Geometry & ML and have a nice research project in mind? 📐🤖 Join us at #LOGML2026 as a mentor! Help a brilliant team turn your project idea into a research success story.

Our Geometric Neural Distance Fields will be presented at #CVPR2026: circle-group.github.io/research/NRMF/. @CVPR

We present Neural Riemannian Motion Fields (NRMF) as a promising paradigm for generative modeling of articulated motion. Unlike various diffusion based models, NRMF is useful both in generation and in deployment as a prior across tasks: circle-group.github.io/research/NRMF/. #CVPR #AI #CVML👇

Months of planning, and it's over in a flash! Seeing #LOGML come to life with such a fantastic group was worth every bit of effort. The collaborative spirit during the projects was exactly what we hoped for. Immense gratitude to all who came and made the week a huge success. 👏

#AI #MachineLearning #Geomtry #Math #NeurIPS #CVPR #ICCV #ECCV #ICML #ICLR #AISTATS #SIGGRAPH

LOGML 2025 is shaping up to be incredible! 🔥 If you're into geometry & ML, this is your chance to dive deep, meet top researchers, and work on projects that matter—all in London 🇬🇧 with an amazing community. Trust me, it's as fun as it is inspiring. Apply now! ⏳👇

@JamesEdwards Good point! Our model isn’t quite ready for super high-res PBR textures due to dataset limits, but we’re working on scaling it up. 🤞 LoD and geometry-driven texture variations are fascinating areas we’re eager to explore with future improvements!

@simo_foti I’m curious to see how this approach can be adapted to a current gen game pipeline that uses PBR materials and textures, starting with a high resolution mesh complete with micro details. We start with this mesh and use it to bake all kinds of useful texture maps for PBR.

🚨 It's confirmed, "UV-free Texture Generation with Denoising and Geodesic Heat Diffusions" just landed at #NeurIPS2024! No more UV map struggles—just point cloud textures & heat diffusion magic. 🔥 Curious? Keep reading. Oh, and definitely turn up the audio 🎧👇

💡Excited by geometry & ML like our organizing committee?💡 Mentor at LOGML 2025 and help shape projects that can become top publications. It’s academic, social, and seriously fun—Plus, it’s in London 🇬🇧 (July 7-11) and we’ve got you covered. Apply now 👇👇👇

@lettercastai Oh nice, thank you for making this alternative podcast! Even if it is probably less engaging than the one I added to the video, it better captures the content of the paper. 🤓 Blending the two versions would be amazing 🙂

@simo_foti Congrats on your #NeurIPS paper! 🎉 We love that there is already an audio version for this paper! Listen to our generation too: lettercast.ai/feeds/neurips_… There are other #NeurIPS2024 papers in the same feed as well :)

@lovespurts I appreciate your insights, but the podcast is just a playful way to present our research. Ultimately, the most important endorsement comes from the scientific community that peer-reviewed our work for NeurlPS, which holds more weight than any podcast commentary.

@simo_foti It's extremely realistic. If someone saw this video without the AI disclaimer they could easily be lead to believe this is an endorsement from a knowledgeable person. Quite an ethically dubious thing to put out, and your response is classic "toxic positivity"

@LEDs_YT We made an AI model that generates textures directly on the surface of 3D objects. The full podcast offers quite a fun simplification if you are interested 🙂!

@simo_foti English?

@lovespurts The podcast is really just a fun, imperfect experiment in making science accessible. It’s not likely to trash any paper—just adds quirky commentary. Honestly, I wish I could have changed a few parts, but it wasn't possible. Let’s keep it light and enjoy the fun side of AI! 😉

@simo_foti So the AI has agency in whether it chooses to praise your paper? Unlikely. And if it trashed your paper, justifiably or not, you wouldn't have posted it, would you?