Domain shift in biomedical imaging is a major unsolved problem; this leaderboard provides a standardized way to measure and compare how robust different AI approaches are to batch effects and experimental variations.
BatchBench is a leaderboard for testing how well AI systems handle images from different experimental batches and sources in biomedical imaging. Built on JUMP-CP, a large cell imaging dataset with significant batch effects, it measures whether AI models trained on one batch can work reliably on new, unseen batches—a critical real-world challenge.