Oliver Hahn

Excited to share our #CVPR2025 highlight on unsupervised panoptic segmentation!

I am presenting AnyUp tomorrow at #ICLR2026 in the Oral session 5E and my poster is P4-3517. Please come by to chat about this project!

📢 Excited to be in Rio at #ICLR2026 for our work Lyra with @xuanchi13. Looking forward to discussing generative world models and grabbing Caipirinhas at Ipanema Beach. Poster: P4-#3208, Thursday 10:30 AM - 1 PM Lyra 1.0: research.nvidia.com/labs/toronto-a… Lyra 2.0: research.nvidia.com/labs/sil/proje…

What if a model could learn dense semantic matches from just a handful of annotated landmarks, while still generalizing to unseen keypoints and categories — and running 10× faster than diffusion-based approaches? MARCO is selected as an Oral at #CVPR2026! A unified model for generalizable semantic correspondence, built on DINOv2⭐️ 👉 Try our model: github.com/visinf/MARCO

📢 Check out Lyra 2.0: large-scale generative worlds with a 3D-consistent video diffusion model. Optionally, feed-forward 3DGS and/or mesh reconstructions. Great work by @TianchangS, @xuanchi13, and the rest of the crew. Grateful for my time with all of these brilliant people 🙏

Today, we released Lyra 2.0, a framework for generating persistent, explorable 3D worlds at scale, from NVIDIA Research. Generating large-scale, complex environments is difficult for AI models. Current models often “forget” what spaces look like and lose track of movement over time, causing objects to shift, blur, or appear inconsistent. This prevents them from creating the reliable 3D environments required for downstream simulations. Lyra 2.0 solves these issues by: ✅ Maintaining per-frame 3D geometry to retrieve past frames and establish spatial correspondences ✅ Using self-augmented training to correct its own temporal drifting. Lyra 2.0 turns an image into a 3D world you can walk through, look back, and drop a robot into for real-time rendering, simulation, and immersive applications. ➡️ Learn more: research.nvidia.com/labs/sil/proje… 📄 Read the paper: arxiv.org/abs/2604.13036

@ClaudiaCuttano Congrats! 🚀

✨ As a first-year PhD student, I used to wonder what it must feel like to have a paper selected as an Oral at #CVPR. Today, I’m experiencing that feeling twice! I’m beyond happy to share that both of my first-author papers have been selected as #Oral at #CVPR2026 🎉

@gabTrivv 🔥

🔥 Can in-context segmentation emerge directly from frozen DINOv3 features? At #CVPR2026, we present INSID3: Training-Free In-Context Segmentation with DINQv3 — a collaboration between PoliTo, TU Darmstadt and TU Munich. A training free approach that generalizes from object-level to part-level and personalized segmentation, across natural, medical, underwater, and aerial domains Check it out: github.com/visinf/INSID3

Congratulations Dr. @sherwinbahmani

@gabTrivv @Meta @p_bojanowski Congrats! 🎊

Excited to share I’ve joined @Meta as a Postdoc on the DINO team 🦖 Grateful to @p_bojanowski for the opportunity, looking forward to what’s ahead!

📢🎓 We have open PostDoc positions in Computer Vision & Machine Learning at @TUDarmstadt and hessian.AI within the Reasonable AI Cluster of Excellence — supervised by @stefanroth, @schaub_simone and many others! Apply here: career.tu-darmstadt.de/tu-darmstadt/j…

📢 NeurIPS 2025 Spotlight 📢 Can we embed motion into image representations? Trained on videos, FlowFeat embeds optical flow into pixel-level representations (up to a linear transform), which results in sharp feature grids, especially for dynamic objects. We demonstrate benefits for ⭐️video object segmentation; ⭐️semantic segmentation; ⭐️and monocular depth. Paper: arxiv.org/abs/2511.07696 Project website: tum-vision.github.io/flowfeat Code and models: github.com/tum-vision/flo… Joint work with Anna Sonnweber and Daniel Cremers @tumcvg and @MunichCenterML. Come by our poster @NeurIPSConf on Thursday (Exhibit Hall C,D,E #4816)!

📢 Join @TUDarmstadt as a PhD/Postdoc in the new project HAICC - Human–AI Collaboration for Cybersecurity! Explore how LLM–based AI agents and humans can jointly analyse security data and rethink cybersecurity architectures - supervised by @IGurevych, @stefanroth, and many more!

✨ We found that #SegmentAnything hides a rich semantic structure, and we show how to unlock it! Our paper SANSA: Unleashing the Hidden Semantics in SAM2 for Few-Shot Segmentation is a #NeurIPS2025 Spotlight. 📍 Come check it out! Poster Friday, 11 a.m. 📄github.com/ClaudiaCuttano…

🔥 Our paper SANSA is a #NeurIPS2025 Spotlight! We turn #SAM2 into a semantic few-shot segmenter for objects and parts, fully promptable (mask · point · box · scribble); only 10M trainable parameters and 5× faster than competitors. Code, models & demo github.com/ClaudiaCuttano… 👇

🔔 AnyUp implementation update: 🔔 ⚡️Our new implementation slashes memory + runtime requirements, with bigger gains on larger images. ✨New model trained on multiple backbones that generalizes even better to unseen features at test time. Check it out: github.com/wimmerth/anyup

@NeurIPSConf is two weeks away! 📢 Stop missing great workshop speakers just because the workshop wasn’t on your radar. Browse them all in one place: robinhesse.github.io/workshop_speak… (also available for @EurIPSConf ) #NeurIPS #neurips2025 #EurIPS

📢🎓 We have open PhD positions in Computer Vision & Machine Learning at @TUDarmstadt and hessian.AI within the Reasonable AI Cluster of Excellence — supervised by @stefanroth, @schaub_simone and many others! Apply here: career.tu-darmstadt.de/tu-darmstadt/j…

🎉 Today, Simon Kiefhber will present our ICCV oral paper on how to make optical flow estimators more efficient (faster inference and lower memory usage) with state-of-the-art accuracy: Talk: Tue 09:30 AM, Kalakaua Ballroom Poster: Tue 11:45 AM, Exhibit Hall I #76

Interested in 3D DINO features from a single image or unsupervised scene understanding?🦖 Come by our SceneDINO poster at NeuSLAM today, 14:15 (Kamehameha II room) or Tue, 15:15 (Exhibit Hall I #627)!🖼️ W/ A. Jevtić, @felixwimbauer @olvr_hhn, C. Rupprecht, @stefanroth, D. Cremers