🚀 Excited to share our new work RefAM: Attention Magnets for Zero-Shot Referral Segmentation, a training-free approach that turns diffusion model attentions into segmentations. By @anna_kukleva_, me, Alessio Tonioni, @ferjadnaeem, @fedassa, @janericlenssen, Bernt Schiele

Need to scale up any vision feature (e.g. DINOv3) to any resolution without retraining? Make sure to check out AnyUp, @wimmer_th's new work!

We don’t have the answers but give you a tool to test them. We welcome feedback and contributions!

SRM results show there’s no single best setup for every dataset or task: *  Which noise schedule, parameterization, or diffusion formulation works best? *  DiT, UNet, U-ViT, or something custom? *  Train from scratch or fine-tune? *  How much to sequentialize and which order?

Just define the variables — elements of a sample that share the same noise level — and the package does the rest. Below are examples for image patches and video frames as variables.

Spatial Reasoners allow to bring (soft) autoregressive diffusion/flow models to a wider range of domains — aiming to facilitate research on denoising models for reasoning and generation tasks over multiple variables.

Poster: #CodeML Workshop · West Meeting Room 211-214 · 2:15 pm on Friday Webpage: https://t.co/l3kdQ5h2kF Github: https://t.co/n6TMwXVOFk Paper:

You can bring our Sudoku solving diffusion models to other domains! If you are at interested and at #ICML2025, come see @bartek_pog and @ChrisWewer's 🌀 Spatial Reasoners package — now released in beta! Here are some examples for images and videos. Links below.

@ChrisWewer @bartek_pog PS: 🌀We recently released spatial-reasoners, a general toolkit to apply SRMs to a wide range of different domains:

@ChrisWewer @bartek_pog We find that model hallucination can be drastically reduced by choosing the right configuration, allowing to significantly increase performance in complex reasoning tasks like solving visual Sudoku.

@ChrisWewer @bartek_pog Our Spatial Reasoning Models allow to explore the space between parallel and autoregressive diffusion models with different methods for choosing generation order. Project Website:

Can diffusion models solve visual Sudoku? If you are at #ICML2025, come to our poster in the Wednesday morning poster session (Poster Session 3 East, Poster 3412) and find out! @ChrisWewer @bartek_pog Bernt Schiele @janericlenssen

📢Is your multi-view generation (MVG) model 3D consistent? Do they produce high-quality and semantically correct novel views? How to fairly compare and make them even better? Introducing MVGBench: a comprehensive benchmark for MVGs, accepted to #ICCV25 @ICCVConference

MET3R quantitatively measures 3D consistency between two images via DUSt3R reconstruction and feature comparison. It does not require camera poses. Code is available for plug-and-play use. We also provide an open source multi-view latent diffusion model for further research!

Project page:

At #CVPR2025 and working on consistency in video and multi-view generative models? Come and visit our poster on Friday afternoon, where I present 𝗠𝗘𝘁𝟯𝗥: 𝗠𝗲𝗮𝘀𝘂𝗿𝗶𝗻𝗴 𝗠𝘂𝗹𝘁𝗶-𝗩𝗶𝗲𝘄 𝗖𝗼𝗻𝘀𝗶𝘀𝘁𝗲𝗻𝗰𝘆 𝗶𝗻 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗲𝗱 𝗜𝗺𝗮𝗴𝗲𝘀

Introducing KumoRFM – the world’s first Relational Foundation Model for enterprise data. Instantly generate accurate predictions like churn, fraud, and recommendations from raw data, with zero-shot predictions and 20x faster results. Try KumoRFM today: https://t.co/h2r8UdtZkk

@ChrisWewer @bartek_pog We also show that good orders can be predicted by uncertainty, which is crucial for the Sudoku task to be solved well.

@ChrisWewer @bartek_pog It allows to explore the amount of (soft) sequentialization and the order of generation, both having significant impact on reasoning quality.

@ChrisWewer @bartek_pog Spatial Reasoning Models (SRMs) are a framework to propagate belief over a set of continuous variables (e.g. image patches) with generative denoising models.