Semi-supervised Instance Segmentation with a Learned Shape Prior
This work addresses the challenge of data annotation in instance segmentation for domains like cell segmentation, offering a semi-supervised approach that reduces reliance on large labeled datasets.
The paper tackles the problem of instance segmentation by proposing a framework that uses a learned shape prior to reduce the need for extensive annotated training data, achieving results comparable to fully supervised methods on two out of three cell segmentation datasets and outperforming pre-trained supervised models with limited domain-specific data on all three datasets.
To date, most instance segmentation approaches are based on supervised learning that requires a considerable amount of annotated object contours as training ground truth. Here, we propose a framework that searches for the target object based on a shape prior. The shape prior model is learned with a variational autoencoder that requires only a very limited amount of training data: In our experiments, a few dozens of object shape patches from the target dataset, as well as purely synthetic shapes, were sufficient to achieve results en par with supervised methods with full access to training data on two out of three cell segmentation datasets. Our method with a synthetic shape prior was superior to pre-trained supervised models with access to limited domain-specific training data on all three datasets. Since the learning of prior models requires shape patches, whether real or synthetic data, we call this framework semi-supervised learning.