Transformers and CNN in analyzing microscopy images

• python 3.7 virtual environment https://docs.hpc.ut.ee/cluster/python_envs/. Install all requirements pip install -r requirements.txt
• Running HPC with gpu: Use TransUNet/start_*.sh scripts.
• To execute job in HPC: sbatch start_*.sh. To monitor the execution tail -f slurm-*.out
• Testing logs (and performance measures DICE/pixel-wise F1, HD95) available in TransUNet/test_log.

---

Branches:

• master - Initial model with cropping
• additional_CNN - additional CNN or ResNet to bottleneck changes
• resnet_skip - additional skip-connection from ResNet hidden features to bottleneck
• transformer-to-skip - moving transformers from encoder to skip-connection.

---

Initial architecture source code repo: https://github.com/Beckschen/TransUNet