#MSCA AI Postdoctoral Researcher at Visual Recognition Group, @CVUTPraha. Photographer. Crossfit freak.
Prague
Joined May 2021
- Tweets460
- Following623
- Followers465
- Likes1.8K
Pinned Tweet
1/n Attention, Please! 🚀
Our work “Revisiting Attentive Probing Through the Lens of Efficiency” has been accepted at #ICLR2026.
We introduce Efficient Probing (EP) — a lightweight, multi-query attentive probing method for frozen encoders.
Paper + code at the end 👇
Bill Psomas retweeted
We are in an era where you can agentify almost anything, and often you get a more powerful system.
This work provides a neat example of that approach applied to optimizers:
1/9 Current optimizers are static during training. What if the optimizer could think and react while the network trains?
We let an agent watch the training dynamics and rewrite the optimizer's source code as training goes: a reactive optimizer.
In every language and image setting we tried that has no tuned recipe, it matches or beats the best recipe one can one-shot, and it guards against loss spikes and overfitting by itself. In the near future, optimization decisions may belong to agents, not humans.
📝 xingjianbai.com/reactive-opt…
💻 Try it on your task: github.com/xingjian-bai/opal
🧵
Bill Psomas retweeted
Creating a good eval is something every researcher should do at least once
There are so many bad evals around. So so many
Bill Psomas retweeted
DINOv2 has never seen a caption, and Qwen3 has never seen an image. We still aligned their embedding spaces without a single image-caption pair.
It even works when the images and the captions come from different datasets.
Project page: dominik-schnaus.github.io/un…
⬇️
Bill Psomas retweeted
Besides scalability, self-supervised learning is attractive because we have a good real-life example of humans/mammals learning strong visual systems.
BUT: they do not learn from shuffled image collections. They learn from (essentially) a single long visual stream.
This is what we tackle in our new @NeurIPSConf paper, led by Ivan, together with @LukasKnobel1 and me. 1/
arxiv.org/abs/2609.40333
Bill Psomas retweeted
Call me old, but I ❤️ walking aimlessly through bookstores and discovering new books.
Bill Psomas retweeted
🤖 🚀 I am opening one fully-funded 2-year postdoc position to explore egocentric 4D scene representation and understanding for robot learning, with fantastic collaborators @FrancisEngelman and @antoniloq
⏰ Deadline: November 23rd, 23:59
🔗 lnkd.in/e4ZW5WrB
🤖 Made with AI
Bill Psomas retweeted
Many people at ECCV 2026 were actually worried about computer vision (and their own) future after the Astra came out.
If you are as well - watch this @dhh talk. That's the only good way to look at things.
youtube.com/watch?v=vDjW_dRy…
Bill Psomas retweeted
Opus 5.5, GPT-6 Sol, and GPT-6 Luna numbers are on the Playground leaderboard
crazy how fast Google lost the computer vision lead
object detection used to be their stronghold, now 3 labs and 4 models sit ahead of them
link: playground.roboflow.com/eval…
Bill Psomas retweeted
Time for a hard cap at 10 paper…? together with submission limits based on accepted work?
This quoted post is unavailable.
🔥 Composed image retrieval, as it should be.
📸 i-CIR (NeurIPS 2025): a photo says which particular object, a text says how it should look.
🔍︎ Explore every composed query in our new dataset browser: vrg.fel.cvut.cz/icir/?v=2
More to come...
#ComputerVision #ImageRetrieval
Bill Psomas retweeted
The #ECCV2026 Malmö 🇸🇪 reception is in full swing!
Bill Psomas retweeted
Happy to co-host the 2nd edition of the FOUND Workshop in @eccvconf.
For this edition, we are heavily emphasizing on Physical AI and World Models, with a specific focus on data curation for real-world embodiment.
Join us on 9th Sept 8:30 to 12:30 at MalmöMässan E1. More info⬇️
Bill Psomas retweeted
Most of what @the_good_ai_lab does can be described as building a connection. A bridge between frontier AI research and humanitarian action, education, and sustainability. 🌱🧵
cv4good.thegoodailab.org w/ @MSF
🎉 Our #ICLR26 paper on efficient probing (EP) now has a home that outlives the paper.
⏱️ Probing Frozen Encoders — a standing ImageNet-1k benchmark for k-NN, linear (LP) and efficient probing (EP).
Built to stay. Issues and PRs welcome 👇
github.com/billpsomas/effici…
🤝 Want your model in the table?
Clone, run the three commands (k-NN, LP, EP), open a PR with one row.
That's it — the leaderboard grows with the field.
🔮 Coming next:
🧠 EP head checkpoints — load & probe, no training 🎨 attention-map viz of EP's queries
🛰️ remote sensing foundation models
🆕 ReImageNet re-eval — ~12% of IN-1k val labels are wrong! (arxiv.org/abs/2608.13783, @klaaracz)
Stay tuned 🚀
