IONIAN at ECCV 2026

The IONIAN team is at the European Conference on Computer Vision (ECCV 2026) in Malmö, Sweden, with 1 publication! 🎉

Our paper: On Test-Time Scaling for Vision-Language Models

Can test-time scaling methods developed for LLMs be directly applied to Large Vision-Language Models? In this work, Fawaz Sammani, Tzoulio Chamiti, and Nikos Deligiannis present the first comprehensive study of test-time scaling for LVLMs — spanning multiple models and sizes, nine test-time scaling methods, and six benchmarks.

Key findings:

• Small, well-performing models benefit the most from test-time scaling, with gains of up to ~30%, often matching or outperforming much larger models,
• LVLMs lose focus when given more compute than necessary,
• Visual information is encoded early in the reasoning chain, after which text-only reasoning dominates and image tokens contribute much less.

📄 Paper: arxiv.org/abs/2606.28864

Tzoulio and Fawaz presented the work at the conference:

Around the conference

Beyond the talks and posters, ECCV was a great chance to connect with the community and spend time together as a team.

See you at the next one! 👋