News
GCPR Best Master's Thesis AwardGCPR Best Master's Thesis Award
September 23, 2026
We are happy to announce that Sunil Khatri, a Master’s student supervised in the SEEALL group, has received the DAGM Best Master’s Thesis Award 2026 as part of the Young Researchers’ Forum. The award was presented at this year’s GCPR and recognizes outstanding Master’s theses in pattern recognition and computer vision. Congratulations, Sunil!
In his thesis, Sunil investigated Visual In-Context Learning (VICL), where vision models adapt to new tasks at test time from just a few example image pairs, without retraining. Instead of following the trend toward ever larger models, he trained a model with only 1 million parameters and around 70,000 training images, and compared it with VICL models up to 7,000× larger. Surprisingly, the tiny model holds its own and even outperforms the large models in several settings. The work shows that current VICL models often fall back on tasks seen during pre-training rather than truly learning from the context, and that common evaluation metrics do not capture this well. The results are available in the paper “Beyond Model Size: Probing the Gaps in Visual in-Context Learning by Training a Tiny Model”. More information about the award is available on the DAGM website.