Sliding Window and Pseudo-labeling techniques for Cellular Segmentation

The Nam Nguyen, Nguyen Hai Minh, Truong Nhat Bui, Tuyen Trong Dam, Hanh Thi Le, Ngoc Anh Nguyen Kha, Duong Le Huy, Anh Nguyen Cong Hoang, Anh Ngoc Nguyen, Le Trung Kien, Duong Hai Nguyen

Keywords: Cell Segmentation

TL;DR: A solution to NeurIPS 2022 Weakly Supervised Cell Segmentation

Abstract:

Cell segmentation is a fundamental task in biomedical image analysis, which
involves the identification and separation of individual cells from microscopy
images. Large-size images and unannotated data are two canailing problems
degrading the performance in cell segmentation. Regarding these issues, we
propose sliding window and pseudo-labeling techniques by conducting several
experiments on different neural architectures. Following this approach, our method
achieves a significant performance improvement and a final result of 0.8097 F1
score on the tuning set and 0.6379 F1 score on the test set of Weakly Supervised
Cell Segmentation in Multi-modality Microscopy challenge hosted at NeurIPS
2022.