Abstract: Cell Segmentation is an initial and fundamental step in biomedical image analysis, which strongly affects the experimental results of this analysis. Recently, deep learning based segmentation methods have shown great power in segmentation accuracy and efficiency. However, these data-driven methods still face many challenges, such as lack of annotations, multi-modality, and complex morphology, where morphological complexity significantly limits model performance. In this paper, we propose a new all-purpose framework with high morphological adaptability for multi-modality cell segmentation, termed Cell Segmenter (CS). For high convex cells with an arbitrary size, the Anchor-based Watershed Framework (AWF) precisely locates well-defined cell centers and generates segmentation based on these markers. For those elongated or non-convex cells, the center-independent segmentation method Omnipose is adopted to obtain satisfying masks. In the inference time, confidence-based quality estimation is conducted on the branch predictions if needed, and then the better result is chosen as the final segmentation. The F1-score of the proposed method reaches 0.8537 on TuningSet and 0.6216 on the final test set of the NeurIPS 2022 Cell Segmentation Challenge.