Rethinking mask heads for partially supervised instance segmentation

We focus on partially supervised instance segmentation where only a subset of categories are mask -annotated (seen) and the model is expected to generalize to unseen categories for which only box anno-tations are provided to eliminate laborious mask annotations. Many recent studies train a class-agnostic segmentation network to distinguish foreground areas in each proposal. However, class-agnostic models behave poorly in complex contexts when the foreground object overlaps with other irreverent objects. Identifying specific object categories is simpler than distinguishing foreground from background since the definition of the foreground is ambiguous even for a human. However, training class-specific model is unfeasible under the partially supervised setting since the mask annotations of unseen categories are absent during training. To overcome this issue, we put forward a teacher-student architecture where the teacher learns general yet comprehensive knowledge and the students, guided by the teacher, delve deeper into specific categories. Concretely, the teacher learns to segment foreground from proposals and the student is devoted to segmenting objects of specific categories. Extensive experiments on the chal-lenging COCO dataset demonstrate our method consistently improve the performance of several recent state-of-the-art methods for the partially setting. Especially, for overlapped objects, our method signifi-cantly outperforms the competitors with a clear margin, demonstrating the superiority of our method.(c) 2022 Elsevier B.V. All rights reserved.

Automated summary

This paper focuses on partially supervised instance segmentation and proposes a teacher-student architecture where the teacher learns general knowledge and the students delve into specific categories. Extensive experiments on the COCO dataset demonstrate the superiority of this method.