• Repo: PictSure on GitHub
• Paper: arXiv:2506.14842
• Website: PictSure.eu

PictSure is a lightweight vision-only in-context learning (ICL) family of models for few-shot image classification (FSIC). It uses pretrained frozen embeddings and outperforms CLIP-based models on out-of-domain tasks like medical imaging.

Contacts: lukas.schiesser@dfki.de & cornelius.wolff@cwi.nl