COOS-7 Accuracy

The COOS-7 Accuracy task aims to predict the subcellular localization of proteins in mouse cells based on a dataset containing 132,209 single-cell images. This task provides a classification setting where four test datasets have varying degrees of covariate shift, ranging from a random subset of the training data to images obtained from re-experimentation using different instruments months later. The goal is to evaluate and improve the model's generalization capability in the face of significant covariate shifts, thereby enhancing the value of computer vision in biomedical applications.