←
FunAI33 posts
FunAI avatar

FunAI

@FunAILab

Research lab led by @y_m_asano at @utn_nuremberg. We conduct fundamental AI research and develop core technology for future Foundation Models.

◷ Joined July 2024▧ 6 Photos and videos
Andrei Bursuc avatar
⟳ FunAI retweeted
Andrei Bursuc@abursuc·Nov 13

Vision Banana finetuned a closed model & shared no code, weights or data. Its report left many design choices open, so we worked them out ourselves. OVB: one open FLUX.2 Klein model for segmentation, depth & normals that still edits images. 🍌Super fun project w/ @FunAILab team

🍌 What if an image generator could see? @GoogleDeepMind's Vision Banana showed it can, but kept data, recipe and weights closed. So we rebuilt it fully open: ✨ FLUX.2 Klein 4B/9B finetune w/ public data only ✨ 73.9 vs Vision Banana’s 69.9 mIoU zero-shot on Cityscapes ✨We release code + 148 checkpoints for the community/ Aaand we innovate: 📈 We develop a new test-time scaling recipe we call “legend-ensembling”: this gives strong gains by applying more compute without retraining anything. 🔗 openvisionbanana.github.io 📄 arxiv.org/abs/2611.15844 A @FunAILab work from A. Jadon, @v_pariza, @FragileGoodwill, @jonaruthardt, @ASophiaKoepke, @abursuc and me.

0:45
Yuki avatar
⟳ FunAI retweeted
Yuki@y_m_asano·Nov 13

🍌 What if an image generator could see? @GoogleDeepMind's Vision Banana showed it can, but kept data, recipe and weights closed. So we rebuilt it fully open: ✨ FLUX.2 Klein 4B/9B finetune w/ public data only ✨ 73.9 vs Vision Banana’s 69.9 mIoU zero-shot on Cityscapes ✨We release code + 148 checkpoints for the community/ Aaand we innovate: 📈 We develop a new test-time scaling recipe we call “legend-ensembling”: this gives strong gains by applying more compute without retraining anything. 🔗 openvisionbanana.github.io 📄 arxiv.org/abs/2611.15844 A @FunAILab work from A. Jadon, @v_pariza, @FragileGoodwill, @jonaruthardt, @ASophiaKoepke, @abursuc and me.

0:45
Yuki avatar
⟳ FunAI retweeted
Yuki@y_m_asano·Oct 7

@FunAILab with me and @dawkopi at @COLM_conf in SF! Seeing lots of cool papers already and stoked for meeting more folks from the LM side these days! Just reach out and let’s grab a coffee ✌️

Post attachment
Yuki avatar
⟳ FunAI retweeted
Yuki@y_m_asano·Sep 10

I am incredibly honoured to receive the ECVA/ @eccvconf Young Researcher Award. ✨🎉I cannot quite believe it, but what I do know is, that this is all thanks to my amazing collaborators, mentors and students and friends. Thank you! You all put the Fun in Fundamental AI. It’s an adventure to be in this field, in this time.

Post attachmentPost attachment
FunAI avatar
FunAI@FunAILab·Sep 4

Our lab regularly hosts visiting PhD students from other labs! It's a great way to exchange insights, expand the network and build collaborations. 🎉

Post attachment
Yuki avatar
⟳ FunAI retweeted
Yuki@y_m_asano·Aug 22

My opposing take: text will get you only so far (maybe far enough for many applications and commercial value), but learning from unlabeled data, in particular videos, will (with the right method/architecture) be the best: huge quantities, physics, geometry, occlusions etc. but: Getting there will take proper research, no winning recipe there just yet

3. When asked about DINO (and SSL in general): "Text is there so why not use it? [...] I think it's an intellectually interesting concept and we should be able to learn good representations without text and it's nice that they got this far, but (not using text) it's unnecessarily making it hard" [3/3]

Post attachment
Ioana avatar
⟳ FunAI retweeted
Ioana@ioanasimioni·Jul 5

(1/n) What if you could teach a vision model about 3D structure without a single 3D label and just a tiny fraction of its pretraining compute? 👁️ That’s 3DPoV. ✨ Heading to #ICML2026 in Seoul 🇰🇷 to share it 👇

Post attachment
Yuki avatar
⟳ FunAI retweeted
Yuki@y_m_asano·Jun 18

[PhD opening🎉] Part of the Bavarian Foundation Model initiative is to also develop and build more radical ideas (besides catching up). Within this, I'm stoked to be involved in an exciting project on tokenizer-free FoMos with @vjhofmann 👇

📢 We're hiring a PhD student in tokenizer-free foundation models! The position will have a double affiliation with @LMU_Muenchen and @utn_nuremberg, co-supervised by @y_m_asano and me. Come work with us in Munich and Nuremberg! 👇

Post attachmentPost attachment
Oishi Deb avatar
⟳ FunAI retweeted
Oishi Deb@DebOishi·Jun 3

Happening Now! Come to VGI workshop @CVPR in Room 703 to listen to the excellent keynote speakers: Robert Geirhos @GoogleDeepMind; @AdtRaghunathan @CSDatCMU; @mattdeitke @Meta; @y_m_asano @funailab; @alexgkendall @wayve_ai; Andrea Vedaldi; Alexei Efros; Kristen Grauman @CVPRConf

Post attachment

Checkout our @CVPR VGI workshop on 3rd Jun 2026 cvpr2026-vgi-workshop.limitl… in Room 703, Led by @HirokatuKataoka with Yoshihiro Fukuhara, @YonglongT, @elliottszwu, Ryousuke Yamada, @jianyuan_wang, Kohsuke Ide, @xianzhengoxford, @Koichi_N_, @yimiche @Oxford_VGG! @CVPRConf #CVPR2026!!

Post attachmentPost attachment
FunAI avatar
FunAI@FunAILab·Apr 30

Come join us!

🎉[openings] I’m hiring postdoctoral researchers to join our @FunAILab at UTN through the Alexander von Humboldt Research Fellowship (@AvHStiftung), via the Henriette Herz Scouting Programme. As a Henriette Herz Scout, I can nominate outstanding international researchers for this fellowship route. I’m especially keen to hear from candidates working on multimodal learning, video and image pretraining, and post-training. Fellows would be hosted in our lab at UTN and work closely with us on these topics. Key requirements: * finished your doctoral studies less than 4 years ago or will finish in the next 6 months * did not live/work in Germany in the last 10 years * applications from female, trans* and/or non-binary candidates are highly encouraged! Interested? Please send a short note with your CV, PhD year, current affiliation, 2–3 key publications, and a few lines on how your work connects. Please share! 🔀

Post attachment
FunAI avatar
FunAI@FunAILab·Apr 10

Sakana means fish in Japanese.

Today we visited Japan's hottest AI startup @SakanaAILabs🎏🇯🇵! We met their research scientists and discussed the implications and impact of some their works like "The AI scientist" and "Continous Thought Machines". We presented our @FunAILab works, "Better Language Models Exhibit Higher Visual Alignment" and "Data Repetition Beats Data Scaling in Long-CoT Supervised Fine-Tuning". Got lots of cool questions and discussions! Thanks, Masanori Suganuma, @_yutaroyamada, @ciaran_regan_

Post attachmentPost attachmentPost attachmentPost attachment