Wouldn't it be great if we could merge the knowledge of 20 specialist models into a single one without losing performance? 💪🏻 Introducing our new ICML paper "Localizing Task Information for Improved Model Merging and Compression". 🎉 📜: arxiv.org/pdf/2405.07813 🧵1/9

SPOT: Self-Training with Patch-Order Permutation for Object-Centric Learning with Autoregressive Transformers by @IoannisKakogeo1@SpyrosGidaris tsiou.karank N. Komodakis tl;dr: improve slot-based autoencoders w/ self-training & patch permutations #CVPR2024 x.com/IoannisKakogeo…

Hi #CVPR2024 people! If you are interested in unsupervised object-centric learning and object segmentation, come and chat with us! You'll find us at poster board 315! 🎉

Day 2 of #IGARSS2024 #Summerschool starts today! From Data to Application 📊➡️📱 Today we start with an in-depth session on machine learning for Earth Observation by tsiou.karank, Bill Psomas, Ioannis Kakogeorgiou🌍🤖 Let's dive in! 🚀💻 #RemoteSensing #Athens #machinelearning

🌊 Exciting news! Our AI framework for tracking global marine pollution, including debris & oil spills, is featured on #NVIDIA's blog! 🚀🌍 By using deep learning & satellite imagery, we boost ocean cleanup efforts. 🌐💡 Read more: developer.nvidia.com/blog/high-tech… #AI NVIDIA AI Developer

Thanks NVIDIA AI Developer for featuring our work on marine pollution detection! 🌊🐬 Excited to leverage advanced AI for a cleaner ocean.

1/n 🚀 Excited to share our latest work: DINO-Foresight, a new framework for predicting the future states of scenes using Vision Foundation Model features! Links to the arXiv and Github 👇

1/n🚀If you’re working on generative image modeling, check out our latest work! We introduce EQ-VAE, a simple yet powerful regularization approach that makes latent representations equivariant to spatial transformations, leading to smoother latents and better generative models.👇

Here's EQ-VAE as well. Always nice when two papers come to the same conclusion independently, that makes the evidence so much stronger! x.com/ThKouz/status/…

EQ-VAE: such a simple & cool trick to regularize multiple kinds of autoencoders: align reconstruction of transformed latents w/ the corresponding transformed inputs. 🚀REPA: 4x training speedup 🚀MaskGIT: 2x training speedup 🚀DiT-XL/2: 7x faster convergence Kudos Thodoris Kouzelis et al.

ILIAS: Instance-Level Image retrieval At Scale Giorgos Kordopatis-Zilos, Vladan Stojnić , Anna Manko, Pavel Šuma, Nikolaos-Antonios Ypsilantis , Nikos Efthymiadis, Zakaria Laskar, Jiří Matas, Ondřej Chum, Giorgos Tolias tl;dr: new retrieval dataset with guaranteed GT. SigLIP rules. arxiv.org/abs/2502.11748 1/

🚗 Ever wondered if an AI model could learn to drive just by watching YouTube? 🎥👀 We trained a 1.2B parameter model on 1,800+ hours of raw driving videos. No labels. No maps. Just pure observation. And it works! 🤯 🧵👇 [1/10]

🧵 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

New blog post: let's talk about latents! sander.ai/2025/04/15/lat…

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)🧵

One weird trick for better diffusion models: concatenate some DINOv2 features to your latent channels! Combining latents with PCA components extracted from DINOv2 features yields faster training and better samples. Also enables a new guidance strategy. Simple and effective!

ReDi was released on Hugging Face A new generative image modeling framework that bridges generation and representation learning

EQ-VAE is accepted at #ICML2025 😁. Grateful to my co-authors for their guidance and collaboration! Ioannis Kakogeorgiou, Spyros Gidaris, Nikos Komodakis.

Behind every great conference is a team of dedicated reviewers. Congratulations to this year’s #CVPR2025 Outstanding Reviewers! cvpr.thecvf.com/Conferences/20…

1/n 🚀New paper out - accepted at #ICCV2025! Introducing DIP: unsupervised post-training that enhances dense features in pretrained ViTs for dense in-context scene understanding Below: Low-shot in-context semantic segmentation examples. DIP features outperform DINOv2!