Bill Psomas

Sleeping while waiting on an “anywhere in the world” paper decision release. #CVPR2026

Attention, Please! Revisiting Attentive Probing Through the Lens of Efficiency As fine-tuning becomes impractical at scale, probing is emerging as the preferred evaluation protocol. However, standard linear probing can understate the capability of models whose pre-training optim...

8/8 Resources 📄

Paper: arxiv.org/abs/2506.10178
Code: github.com/billpsomas/e...

Joint work with: Dionysis Christopoulos,@eirinibaltzi.bsky.social,@ikakogeorgiou.bsky.social, @tim-arav.bsky.social,Nikos Komodakis,Konstantinos Karantzalos,Yannis Avrithis,@gtolias.bsky.social.

See you @ ICLR 2026🇧🇷

7/n Take-home messages 💡

EP:
- Plug-and-play.
- Compatible with all pre-training families.
- Unlocks the potential of encoders optimized for local representations.
- Complementary with PEFT.
- Better to have it, than not to have it. 👀

6/n EP + PEFT = 🔥

- EP captures information that LoRA alone does not, and vice versa.
- LoRA+EP improves over both pure EP and pure LoRA.

📌 Example: a LoRA+EP configuration with 250K params reaches 72%, 4.3% above linear probing (67.7%), while using over 3× fewer parameters.

5/n Interpretability 🔍

- EP queries specialize in distinct spatial regions.
- Attention maps are complementary.
- Semantic correspondences emerge (e.g. tails, feet).
- Verified quantitatively too.

4/n Designed for local representations🧩

📊 Across ImageNet-1K:

- Consistent gains over k-NN and Linear Probing (LP).
- Particularly strong improvements for MIM, VL, and generative.
- Minimal overhead.

3/n Core observation ⚙️

Prior attentive probing uses redundant projections.

🔍 Introducing Efficient Probing (EP):

📌 Multi-query cross-attention.
🔌 Plug-and-play on top of frozen encoders.
💸 Lightweight and parameter-efficient.

2/n Why revisit probing? 🤔

- Linear probing underestimates encoders optimizing local representations.
- Full fine-tuning is costly at scale.
- Attentive probing helps, yet methods are over-parametrized and not well-studied.

👉 Can we get attention benefits without that much overhead?

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 👇

BBoxMaskPose v2: Expanding Mutual Conditioning to 3D

Miroslav Purkrabek, Constantin Kolomiiets, Jiri Matas

tl;dr: sota in human pose estimations, especially for the hard cases
arxiv.org/abs/2601.15200

Would love to try

Best promo anyone could make for this position 👏🏾🏰 And, amazingly, everything said is true 🎆

Postdoctoral research position in Instance-level visual generation Czech Technical University in Prague (CTU) offers a fellowship program, the CTU Global Postdoc Fellowship. This new and attractive two-year fellowship-program offers excellent researchers who have rec...

I have an opening for a two years post-doc position on instance-level (personalized) visual generation. Eligibility: (i) <=7 years from Ph.D. (ii) studies or 1 year outside of Czechia (ii) >=3 journal with IF or CORE A*/A conference papers. Deadline: 15 Feb.
Details: www.euraxess.cz/jobs/399390

🚀New task: Instance-level Image+Text→Image Retrieval

🔎Given a query image + an edit (“during night”), retrieve the same specific instance after the change — not just any similar object.

🛢New dataset on HF: i-CIR huggingface.co/datasets/bil...

🔥Download, run, and share results!

12/12 Joint work with Giorgos Petsangourakis, Christos Sgouropoulos, Theodoros Giannakopoulos, Giorgos Sfikas, @ikakogeorgiou.bsky.social.

REGLUE Your Latents with Global and Local Semantics for Entangled Diffusion Latent diffusion models (LDMs) achieve state-of-the-art image synthesis, yet their reconstruction-style denoising objective provides only indirect semantic supervision: high-level semantics emerge slo...

11/n Summary🏁

REGLUE shows that the way we leverage VFM semantics matters for diffusion. Combining compact local semantics with global context yields faster convergence and state-of-the-art image generation.

📄arXiv: arxiv.org/abs/2512.16636
💻Project: reglueyourlatents.github.io

10/n Faster convergence🔥

REGLUE (SiT-B/2) achieves 12.9 and 28.7 FID at 400K iterations in conditional and unconditional generation, respectively, outperforming REPA, ReDi, and REG. REGLUE (SiT-XL/2) matches 1M-step SOTA performance in just 700k iterations (~30% fewer steps).

9/n Alignment effects ⚓

External alignment complements joint modeling, but its benefits depend on the signal. Local alignment yields consistent gains, whereas global-only alignment can degrade performance. Spatial joint modeling remains the primary driver.

8/n Local > Global Semantics🧩

Our analysis shows that jointly modeling with patch-level semantics drives most gains. The global [CLS] helps, but fine-grained spatial features deliver a strongly larger FID improvement, highlighting the importance of local structure for diffusion.

7/n Semantic preservation under compression📉

Do compressed patch features retain VFM semantics?

Points show frozen compressed DINOv2 semantics (x: ImageNet top-1 / Cityscapes mIoU) vs SiT-B generation quality (y: ImageNet FID) when trained on VAE latents + compressed features.

6/n Non-linear compression matters 💎

Linear PCA can limit patch-level semantics (e.g., ReDi). We introduce a lightweight non-linear semantic compressor that aggregates multi-layer VFM features into a compact, semantics-preserving space, boosting quality (21.4 → 13.3 FID).

5/n Our method 🧠

REGLUE puts these into one unified model and jointly models:

1️⃣ VAE latents (pixels)
2️⃣ local semantics (compressed patch features)
3️⃣ global [CLS] (concept)
➕ alignment loss as a complementary auxiliary boost.

4/n Main insight 💡

Jointly modeling compressed patch-level semantics ➕ VAE latents provides spatial guidance and yields larger gains than alignment-only (REPA) or global-only (REG).

Alignment loss and a global [CLS] token stay complementary, orthogonal signals.

3/n Key design choice 🧩 Compact spatial semantics matter!

To leverage VFMs effectively, diffusion should jointly model VAE latents with multi-layer VFM spatial (patch-level) semantics, via a compact, non-linearly compressed representation.

2/n More semantics are needed! ➕

Existing joint modeling and external alignment approaches (e.g., REPA, REG) inject only a “narrow slice” of VFM features into diffusion. We argue richer semantics are needed to unlock their full potential.

1/n REGLUE Your Latents! 🚀

We introduce REGLUE: a unified framework that entangles VAE latents ➕ Global ➕ Local semantics for faster, higher-fidelity image generation.

Links (paper + code) at the end👇

✨This NASA/ESA/CSA #Webb Space Telescope image shows the Westerlund 1 cluster, a group of massive stars located in our galaxy which are nearly as bright as a million Suns each. 🌌🎄🧪🔭

🔗 esa.int/ESA_Multimedia/Images/2024/10/The_exotic_stellar_population_of_Westerlund_1

Evaluate your policy on REALM - our new real-to-sim validated benchmark for generalization in robot manipulation!

REALM is a good real-world proxy, evidenced by the high correlation with real-world evals.

ALSO: Follow the rising star @msedlacek.bsky.social and ask him any questions about REALM!

billpsomas/icir · Datasets at Hugging Face We’re on a journey to advance and democratize artificial intelligence through open source and open science.

📣 i-CIR dataset (NeurIPS 25) is now on
@hf.co.

🚀Easier download + better discoverability + WebDataset shards for large-scale use (~750K images).

🤗 Grab it here: huggingface.co/datasets/bil...

#computervision #retrieval #datasets #huggingface #NeurIPS

Yejin Choi on whether small companies or Labs can compete with big ones on AI and LLMS. Neurips 2025 invited talk.