Very excited to share the outcome of my internship @Google where we propose a new elastic visual tokenization architecture: PARCEL: Pool-Anchored Resampling with Conditioned Elastic Queries for Efficient Vision-Language Understanding arxiv.org/abs/2605.30126 Thread below 🧵👇

Excited and deeply honored to receive the #3D Outstanding Doctoral Dissertation Award! 🏆 Huge thanks to the awards committee and to everyone making 3DV such a blast. And I couldn't be happier to share this recognition with my long-time friend and colleague @songyoupeng 🙌✨

if you are at 3DV this week, don't miss Nina's poster on mono2stereo video conversion

Hiring: Student Researcher @ Google Zurich! 🇨🇭 Looking for an EMEA-based PhD student to work on: Efficient 3DGS / Feedforward Models / 3D GenAI. Start: June 2026 or earlier. If interested, please reach out with your CV to: sr-zurich-3d@google.com

📢 Exciting news: The 1st GenRecon3D is coming to #CVPR2026! 🚀 We are exploring how to reconstruct more than what is directly visible using generative priors. Fantastic speaker lineup coming soon. 📍 Website: genrecon3d.github.io/ #CVPR #GenAI #3DVision

This work was developed during my internship at @GoogleResearch, together with Johanna Wald, @HideMatsu82, @Pedro_He_Ca, Timo Ropinski, and @fedassa.

The next OpenSUN3D workshop and challenge will be held @CVPR 🚀 ✨ #CVPR2026 Website: opensun3d.github.io/ Paper and challenge deadlines coming soon 🏆🗒️

@fedassa at NeuSLAM workshop (304A) #ICCV2025

📄 Paper: m-niemeyer.github.io/nexf/st… 🌐 Project page: m-niemeyer.github.io/nexf/in… This is joint work with Fabian Manhardt, Marie-Julie Rakotosaona, Michael Oechsle, Christina Tsalicoglou, Keisuke Tateno, @jon_barron , @fedassa.

#SG2RL returns to @ICCVConference 2025 at Honolulu, Hawai'i 🌴☀️ We call for full papers and extended abstracts on all topics around graphs and scene graphs. Deadline: 26 June (23:59PT) Website: sites.google.com/view/sg2rl With our amazing co-organizers: @eadeli @ir0armeni @fedassa

Huge thanks to the amazing team: @Weiyu_Liu_ (co-lead), Siyi Gu, @dill_pkl , Goutam Bhat, @fedassa , @ManlingLi_ , @nickhaber , @jiajunwu_cs 🌐Project site: ai.stanford.edu/~sunfanyun/l… 💻 Code (we plan to open-source everything): github.com/sunfanyunn/Layout… n/n

Excited to see our paper "Tokenformer: Rethinking transformer scaling with tokenized model parameters" accepted as a spotlight at #ICLR2025 ! Hope our idea of ​​tokenizing everything can inspire the future of AI. Paper: arxiv.org/abs/2410.23168 Code: github.com/Haiyang-W/TokenFo…

Joint work with Fan Yue, @ferjadnaeem, @xyongqin, @janericlenssen, Liwei Wang, @fedassa, and Bernt Schiele. Thanks to everyone on that amazing team.😉😀

Google Internship call: we are looking for a PhD student in the area of object/scene understanding and VLMs to join our Google team in Zurich next summer. If you have applied for the 2025 Google Internship Call already and are interested to know more, ping me!

🚀 Excited to introduce UIP2P: Unsupervised Instruction-Based Image Editing via Cycle Edit Consistency (CEC)! TL;DR: Scalable instruction-based image editing without paired image data. Works on real-world datasets with robust, reversible edits! Thanks to Alessio Tonioni, @xyongqin, @THofmann2017, @fedassa!

Get ready for the next @CVPR workshop on OpenWorld 3D Scene Understanding ➡️ opensun3d.github.io We will be hosting: - prized challenge 🏆 (see scenefun3d.github.io) - paper track 🗞️ - exciting keynote speakers  👩‍🏫 #CVPR2025

SuperGSeg: Open-Vocabulary 3D Segmentation with Structured Super-Gaussians Contributions: • We propose SuperGSeg: a 3D segmentation method with neural Gaussians, designed to learn hierarchical instance segmentation features from 2D foundation models. • We introduce the concept of Super-Gaussian, a novel representation that integrates hierarchical instance segmentation features, enabling the embedding of high-dimensional language features. This approach addresses previously unfeasible challenges in representing complex scenes with rich semantic details. • Extensive experiments on the LERF-OVS and ScanNet datasets demonstrate the effectiveness of the proposed method, achieving significant improvements in open-vocabulary 3D object-level and scene-level semantic segmentation. It shows particular strength in capturing fine-grained scene details and dense pixel semantic segmentation tasks for the first time.

Excited to share LoRACLR, a new way to compose from multiple LoRA models. A collaboration with Google Research!

🚀 Excited to share our preprint LoRACLR! TL;DR: LoRACLR merges multiple LoRA models into a unified diffusion model for seamless, high-fidelity multi-concept image synthesis with minimal interference. Thanks to @THofmann2017, @fedassa, and @PINguAR! 🙌

#AndroidXR is officially here! 🎉 Thrilled to see the developer excitement and passion to build immersive experiences. Huge thanks to the @GoogleARVR team and our incredible partners @SamsungMobile and @Qualcomm and more. ➡️ blog.google/products/android…