Vincent Lepetit

Perplexity Cannot Always Tell Right from Wrong Perplexity -- a function measuring a model's overall level of "surprise" when encountering a particular output -- has gained significant traction in recent years, both as a loss function and as a simp...

Bluesky doesn’t really shine yet as a place to discover papers, especially at the intersection of biology and ML. To help a little, I’m going to start sharing papers I find interesting. To kick things off, here’s one from @petar-v.bsky.social and colleagues at DeepMind
arxiv.org/abs/2601.22950

We have to move all major conferences out of the USA. Nobody wants to get murdered by a militia, ergo no one will attend these conferences. I'm looking at you, CVF...

Latest details about the OpenReview leak by #ICLR2026

blog.iclr.cc/2025/12/03/i...

We introduce MIRO: a new paradigm for T2I model alignment integrating reward conditioning into pretraining, eliminating the need for separate fine-tuning/RL stages. This single-stage approach offers unprecedented efficiency and control.

- 19x faster convergence ⚡
- 370x less FLOPS than FLUX-dev 📉

Did you open-source your #ICCV2025 works?

As a PyTorch Ambassador, I would like to write an article to introduce open-sourced ICCV 2025 works (including workshops and demos) for promoting open-source/science + PyTorch

If interested, share your work via the form in my reply 👇

Familiar names among #ICCV2025 Outstanding Reviewers from our team 😇
Antoine Guédon @antoine-guedon.bsky.social
Sinisa Stekovic
Renaud Marlet
👏
@iccv.bsky.social
iccv.thecvf.com/Conferences/...

Today is Antoine Guedon's PhD! Already pretty cool visuals right at the start.

And lastly, if @neuripsconf.bsky.social would choose to reverse the decisions on the papers affected by space constraints, we would be happy and able to accommodate their presentation

This post from Terence Tao, on the threats to research and science, should be broadly shared wherever you can. It is measured and clear: newsletter.ofthebrave.org/p/im-an-awar...

I am not (very) grumpy (yet)

hey, as long as they cite us, they can butcher the typography as they want

I love the #eurips initiative! But to live up to its potential, it should be accepted an official #neurips conference location (similar to Mexico City) and not just an addon! Then we would save C02, rather than adding to it!!! What can we do to achieve this?

Probably because #NeurIPS2025 reviews were out, I saw some discussions on author-reviewer duties in ML/CV/NLP communities and complaints based on some common misinformation

Currently, CVPR, ICCV, and WACV require all *qualified* authors to serve as reviewers, NOT all authors

We've released code for integrating Gaussian Splatting with PyTorchGeoNodes — enabling procedural Gaussian Splatting!
This allows interpretable 3D reconstructions with better handling of occlusion, noise, and symmetry. Edit splats by tweaking program parameters.

vevenom.github.io/pytorchgeono...

I am heartbroken that I am not at the conference, but seeing what the government is doing to its people and the world, I simply couldn't go there.

I think we did a pretty good job with my amazing #CVPR2025 fellow PCs. With a lot of efforts, and thanks to the hard work of reviewers, ACs, and SACs, the review quality increased while we had to handle a steadily increasing number of submissions.

The #CVPR2025 main conference open access proceedings is up:

openaccess.thecvf.com/CVPR2025

Workshop papers will be posted shortly. Stay tuned...

Imagine retreat 2025 in progress ☀️🏖️

Got some recent research related to 6D Object Pose Estimation? Want to present it at #ICCV2025 in Hawaii? 🌴

Then submit and present at the 𝟏𝟎𝐭𝐡 𝐈𝐧𝐭𝐞𝐫𝐧𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐖𝐨𝐫𝐤𝐬𝐡𝐨𝐩 𝐨𝐧 𝐑𝐞𝐜𝐨𝐯𝐞𝐫𝐢𝐧𝐠 𝟔𝐃 𝐎𝐛𝐣𝐞𝐜𝐭 𝐏𝐨𝐬𝐞 (𝐑𝟔𝐃).

Paper deadlines: Jun 30 (in-proceedings), Aug 29 (non-proceedings)

Philip Isola (seen wearing a blue shirt), who is the CVPR 2025 Program Co-Chair, says "One of the big trends this year in commercial chatbots is they've become multimodal; they now analyze and generate not just text but also images and sometime videos. On the horizon is the ability to generate complete interactive worlds. The image, video, and world synthesis methods being presented at CVPR are paving the way toward this kind of technology." On the bottom right there's a logo of the CVF/ IEEE Computer Society.

#CVPR2025 had a 13% rise in paper submissions--13,008 papers, 40,000+ authors. The technical program chairs say it’s courtesy of the #AI explosion & emerging image, video, and world synthesis methods, among other trends. cvpr.thecvf.com/Conferences/...

#CVPR2025 Sun June 15 (PM)
🧩 RUBIK: A Structured Benchmark for Image Matching across Geometric Challenges
@thibaut_loiseau @GBourmaud
@thibautloiseau.bsky.social @gbourmaud.bsky.social
📄 pdf: arxiv.org/abs/2502.19955

#CVPR2025 Fri June 13 (PM) ✨ Highlight
🍵 MAtCha Gaussians: Atlas of Charts for High-Quality Geometry and Photorealism From Sparse Views
@antoine-guedon.bsky.social @kyotovision.bsky.social
📄 pdf: arxiv.org/abs/2412.06767
🌐 webpage: anttwo.github.io/matcha/

#CVPR2025 Sat June 14 (PM)
🛠️ PyTorchGeoNodes: Enabling Differentiable Shape Programs for 3D Shape Reconstruction
@sinisastekovic.bsky.social Arslan Artykov Stefan Ainetter Mattia D'Urso Friedrich Fraundorfer
📄 pdf: arxiv.org/abs/2404.10620
🌐 webpage: vevenom.github.io/pytorchgeono...

Looking forward to #CVPR2025! We will present the following papers:

Leveraging Automatic CAD Annotations for Supervised Learning in 3D Scene Understanding

Yuchen Rao, Stefan Ainetter, Sinisa Stekovic, @vincentlepetit.bsky.social , Friedrich Fraundorfer

tl;dr: in title
arxiv.org/abs/2504.13580

Reminder! The deadline for #CVPR2025 Social Events proposals is fast approaching. If you are planning to submit, make sure you do not miss the submission closing time!

We're hiring! IMAGINE @ École des Ponts (Paris area) is opening a 4-year "CV for X" researcher position:
– competitive salary
– no teaching load
– starting pkg ≈ 2 PhDs
– goal: impactful core AI + X (climate, biodiversity, robotics...)
Apply by May 31: imagine-lab.enpc.fr/wp-content/u...

The BOP report of 2024 is out: arxiv.org/abs/2504.02812

BOP is a benchmark for 6D object pose estimation, with a new challenge organised yearly.

Community: Model-based 6D localisation? Pff, easy-peasy. 💪
BOP: Ok. Model-free 6D detection on the datasets below 👇
Community: 🙈 (0 entries submitted)

There is some discussion of having a satellite location for NEURIPS 25 outside the US but in a similar time zone. Folks are studying the feasibility and trying to guess how many people would go there