Twitxy Saved

Cees Snoek

@cgmsnoek

Head of Video & Image Sense Lab | University of Amsterdam | Scientific Director Amsterdam AI
Mostly Amsterdam CeesSnoek.info Joined August 2008
1.6K Following 3.9K Followers 3.3K Posts

Latest posts 5

Pinned post

Cees Snoek @cgmsnoek
๐Ÿ‘Š BAM. We are releasing NeoBabel. A multlingual open foundation model for visual generation. It sets a new Pareto frontier in performance, efficiency and inclusivity, natively supporting English, Chinese, Dutch, French, Hindi, and Persian. ๐ŸŒ

Joint work with @Cohere_Labs ๐Ÿซถ
Mohammad Mahdi (Medi) Derakhshani Verified account @mmderakhshani Jul 9, 2025
T1/11
Text-to-image AI is amazing, but what if you don't speak English? Most are English-centric, creating barriers for billions.
We introduce NeoBabel, a new open model for visual generation that natively understands 6 languages, breaking down the walls of digital inequity. ๐ŸŒ๐Ÿ—ฃ๏ธ
View quoted post on X
0 replies 6 reposts 38 likes 3.2K views View on X
Cees Snoek @cgmsnoek
What if the best image #tokenizer isn't the one that preserves the most pixels, but the one that knows what matters? ๐Ÿค” Meet GIST by @danilodegoede
Danilo de Goede @danilodegoede Oct 21
How many tokens does it take to capture the gist of an image?

Introducing GIST: an image tokenizer that learns to preserve what matters and chooses how many tokens each image needs, in one forward pass.

Now on arXiv ๐ŸŽ‰๐ŸŽ‰
arxiv.org/abs/2610.15515

Demo: ddgoede.github.io/gist-tokenizer/
View quoted post on X
Cees Snoek @cgmsnoek
๐Ÿช— Train once, rank once, adapt to any compute budget: meet the secretly elastic Transformer by @iWSimoncini
Walter Simoncini Verified account @iWSimoncini Oct 15
๐Ÿช— Your Transformer Is Secretly Elastic ๐Ÿงต๐Ÿ‘‡

Transformer families come in a few fixed sizes, so you commit to one for your hardware and latency constraints, even as query loads vary. What if a single pre-trained model could serve every budget?

We learn one importance ranking over every MLP neuron and attention head, once, with frozen weights. At inference time, the model can then be pruned to any computational budget by dropping its least important units, without retraining.

Across 27 vision and text encoders spanning 22M to 8B parameters, performance degrades smoothly up to 60% sparsity. For example, pruning DINOv3 ViT-H+/16 to 50% sparsity removes 427M parameters, for only a 0.8% drop in ImageNet-1k linear probing.

Super fun project in collaboration with Dheeraj Varghese (@dhevarghese), supervised by Gertjan J. Burghouts (@gjburghouts) and Cees Snoek (@cgmsnoek)

๐ŸŒ Project page: secretly-elastic.ashita.nl
View quoted post on X
0 replies 0 reposts 11 likes 1.1K views View on X
Cees Snoek @cgmsnoek
๐ŸŽฎ๐Ÿค– What happens when we take todayโ€™s AI agents out of static benchmarks and drop them into thousands of unfamiliar visual worlds?
Dheeraj Varghese Verified account @dhevarghese Oct 9
Ever wondered whether AI agents could play thousands of video games on a supercomputer? We did. ๐ŸŽฎ

Introducing PlaySuite: 5,734 games, 14 open models, and a large-scale benchmark for interactive visual intelligence.

Here's what we built, and what we learned. ๐Ÿงต๐Ÿ‘‡
View quoted post on X