Yeast Cell Segmentation with Generalist Deep Neural Network Mediar-Former

Group members: Yanruiqi Yang(#346510), Yasmine Chaker(#), Ameer Elkhayat(#)

Data preparation:

1. Training Dataset: 906 yeast cell microscopy images with labels, provided by Prof. Sahand Jamal Rahi (Laboratory of the Physics of Biological Systems, LPBS), in the format of .tif;
2. Evaluation Dataset: The Yeast Image Toolkit (Dataset 1, Dataset 3)

Model:

• Pretrained model: Mediar-Former is the "1st winner" in the NeurIPS-2022 Cell Segmentation Challenge
• Predictor: Ensembled with the model and fine-tuned generalist cell segmentation model

Environment Setup:

Training:

Evaluation:

Conclusion:

Reference:

1. MEDIAR: Harmony of Data-Centric and Model-Centric for Multi-Modality Microscopy
2. YeastNet: Deep-Learning-Enabled Accurate Segmentation of Budding Yeast Cells in Bright-Field Microscopy