Gilles Puy (@gillespuy) profile picture
Paris, France
Joined October 2011
In ReDi, we showed that jointly modeling VAE latents + DINO features boosts diffusion. CoReDi goes one step further: the feature space is no longer fixed — it’s learned. → Better joint modeling → 90× faster than SiT → 13× faster than REPA nitter.cf/ThKouz/status/20472669…
1/n Introducing CoReDi: Coevolving Representations for Joint Image–Feature Diffusion Joint diffusion boosts image generation by injecting semantic features. But one assumption goes unquestioned: the feature space is fixed. What if it was learned instead? 🧵👇
12
57
6,467
1/n New paper - V-GIFT 🎁 Self-supervised tasks like rotation prediction or colorization were big in 2018. Do they still matter? Yes. We turn them into visual instruction tuning data for MLLMs. Result: models rely more on the image and perform better on vision tasks 👀
3
22
5
85
12,843
Gilles Puy retweeted
[new CVPR'26 paper] 🔄 SSL works great when you have tons of data. But in 3D… we don’t. High-quality 3D scans are expensive, slow, and hard to scale. So what if we could pretrain 3D models without any real 3D scans? 1/
3
30
5
215
28,620
Gilles Puy retweeted
DrivoR made it to #CVPR2026! Happy for the team and grateful to the reviewers+AC for the feedback. Code and weights are up and ready for test drives.
1/🧵 Q: Can we have both a simple and SOTA architecture in autonomous driving? R: Yes! 😍 Introducing Driving on Registers (DrivoR): a pure Transformer backbone that achieves SOTA results in NAVSIM v1 / v2 and closed-loop HUGSIM evaluation. Here is how 👇
4
23
2,583
1/ 📢 New preprint: Test-Time Conditioning with Representation-Aligned Visual Features Introducing REPA-G — a framework for controllable image generation at inference time using aligned visual features. 📄 arxiv.org/pdf/2602.03753 💻 github.com/valeoai/REPA-G 🧵👇
1
19
1
90
5,342
1/🧵 Q: Can we have both a simple and SOTA architecture in autonomous driving? R: Yes! 😍 Introducing Driving on Registers (DrivoR): a pure Transformer backbone that achieves SOTA results in NAVSIM v1 / v2 and closed-loop HUGSIM evaluation. Here is how 👇
1
12
4
28
12,384
Gilles Puy retweeted
🏠 Introducing CASA: a new way to input visual information into LLMs. The current default to do that is by inserting image tokens into the text stream, but when using many images in long conversations, this floods the context window and is thus impractical for streaming inputs.🧵
2
48
3
254
22,820
Gilles Puy retweeted
The team is at @BMVCconf to present three papers tackling challenges in 3D vision! 🚀 We are presenting new work on: ✨ Diffusion for LiDAR point-clouds 🌙 Depth estimation with light enhancement 🔄 Multimodal distillation for 3D semantic segmentation 👇 #BMVC2025
1
2
1
10
4,170
🚗🌐 Working on domain adaptation for 3D point clouds / LiDAR? We'll present MuDDoS at BMVC: a method that boosts multimodal distillation for 3D semantic segmentation under domain shift. 📍 BMVC 🕚 Monday, Poster Session 1: Multimodal Learning (11:00–12:30) 📌 Hadfield Hall#859
2
3
2
11
1,346
Gilles Puy retweeted
Our recent research will be presented at #ICCV2025 @ICCVConference! We’ll present 5 papers about: 💡 self-supervised & representation learning 🌍 3D occupancy & multi-sensor perception 🧩 open-vocabulary segmentation 🧠 multimodal LLMs & explainability valeoai.github.io/posts/iccv…
1
6
1
20
1,933
Gilles Puy retweeted
The PhD graduation season in the team goes on! Today Corentin Sautier is defending his PhD on "Learning Actionable LiDAR Representations without Annotations". Good luck! 🚀
Another great event for @valeoai: a PhD defense of Corentin Sautier. His thesis «Learning Actionable LiDAR Representations w/o Annotations» covers the papers BEVContrast (learning self-sup LiDAR features), SLidR, ScaLR (distillation), UNIT and Alpine (solving tasks w/o labels).
2
3
16
1,288
Gilles Puy retweeted
It’s PhD graduation season in the team! Today, @Bjoern_Michele is defending his PhD on "Domain Adaptation for 3D Data" Best of luck! 🚀
1
7
21
1,294
Update: ResearchGate has investigated the case, and, as far as I can see, all the suspicious papers (~200) have now been removed. Many thanks to the @ResearchGate team!
Discovered that our RangeViT paper keeps being cited in what might be LLM-generated papers. Number of citations increased rapidly in the last weeks. Too good to be true. Papers popped up on different platforms, but mainly on ResearchGate with ~80 papers in just 3 weeks. [1/]
1
3
638
ReDi has been accepted at #NeurIPS2025 as a Spotlight! Huge thanks to @K_Sta8is, @IoannisKakogeo1, @SpyrosGidaris, and Nikos Komodakis for their guidance and collaboration! This was such a fun project to work on.
1/n Introducing ReDi (Representation Diffusion): a new generative approach that leverages a diffusion model to jointly capture – Low-level image details (via VAE latents) – High-level semantic features (via DINOv2)🧵
3
18
142
10,741
Discovered that our RangeViT paper keeps being cited in what might be LLM-generated papers. Number of citations increased rapidly in the last weeks. Too good to be true. Papers popped up on different platforms, but mainly on ResearchGate with ~80 papers in just 3 weeks. [1/]
1
2
2
6
2,474
But also: (d) Reference lists are very similar across papers. (f) All papers cover almost identical topics (segmentation/detection/recognition in medical imaging), far from the topic of our work. (e) Our paper appears in the bibliography but I never saw it cited in the text. [3/]
1
152
The issue was reported yesterday to @ResearchGate . By the time I created the report, Google Scholar indexed about 30 other additional papers. Curious to know if you’ve seen similar suspicious citations with your papers and what you did about it. [4/]
126
Gilles Puy retweeted
Meet MoshiVis🎙️🖼️, the first open-source real-time speech model that can talk about images! It sees, understands, and talks about images — naturally, and out loud. Voice interaction with a compact model endowed with visual understanding opens up new applications, from audio description for the visual impaired to visual access to information. Try it out 👉 vis.moshi.chat/ Blog post 👉 kyutai.org/moshivis
8
52
12
246
48,026
🧵 Excited to share our latest work: FUTURIST - A unified transformer architecture for multimodal semantic future prediction, is accepted to #CVPR2025 ! Here's how it works (1/n) 👇 Links to the arxiv and github below
3
23
3
100
10,004