EmbedTrack—Simultaneous Cell Segmentation and Tracking Through Learning Offsets and Clustering Bandwidths

To shed light on the processes driving cell migration, a systematic analysis of behavior is required. Since manual hundreds or even thousands cells infeasible, automated approaches for segmentation and tracking are needed. While task deep learning has become standard, there few simultaneous using learning. Here, we present EmbedTrack, single convolutional neural network which predicts human comprehensible embeddings. As embeddings, offsets pixels to their center bandwidths learned processed in subsequent clustering step generate an instance link segmented instances over time. We benchmark our approach nine 2D data sets from Cell Tracking Challenge, where performs seven out within top 3 contestants including three 1 performances. The source code publicly available at https://git.scc.kit.edu/kit-loe-ge/embedtrack.