Cross-Domain Microscopy Cell Counting By Disentangled Transfer Learning

Microscopy images from different imaging conditions, organs, and tissues often have numerous cells with various shapes on a range of backgrounds. As result, designing deep learning model to count in source domain becomes precarious when transferring them new target domain. To address this issue, manual annotation costs are typically the norm training learning-based cell counting models across domains. In paper, we propose cross-domain approach that requires only weak human efforts. Initially, implement network disentangles domain-specific knowledge domain-agnostic images, where they pertain creation style density maps, respectively. We then devise an image synthesis technique capable generating massive synthetic founded few target-domain been labeled. Finally, use public dataset consisting as domain, no cost is present, train our network; subsequently, transfer real images. By progressively refining trained using synthesized several annotated ones, proposed method achieves good performance compared state-of-the-art techniques rely fully evaluated efficacy two datasets actual microscopy cells, demonstrating feasibility requiring annotations