A researcher asks how academic computer vision can keep up with large models

Writing on September 9 ahead of ECCV 2026, a researcher uses their VIGA paper to question how academics can make an impact as large models improve. They say an initial rejection significantly delayed publication, and that VIGA was later surpassed by people using Claude Code. They describe GPT-6 Astra as outperforming all previous results. Their broader concern is the research cycle itself: building on existing literature, developing an idea and publishing it can, they argue, leave conference papers two years out of date.