MASK2TASKS: LEVERAGING SEGMENTATION TO ENHANCE CLASSIFICATION PERFORMANCE IN HISTOPATHOLOGICAL COLORECTAL IMAGES

ABSTRACT

In this paper, we introduced Mask2Tasks, a multi-task learning architecture for colorectal image classification and segmentation, and explored a two-stage training strategy. Through experimentation, we showed that our approach outperforms individual task models and the conventional uniform weighted multi-task learning approach.

Note: Zip file named 'total_notebooks' includes notebook for benchmark models, multi-task learning models with uniform weight, and Mask2task.