At the beginning of the project, we based the entire pipeline on the DeepCell image, allowing us to investigate the implementation of DeepCell in greater detail. However, we are now trying to utilize a GPU to optimize the pipeline, and the DeepCell image has become a pain point due to the use of Python 3.8 by TensorFlow.
The UMAP construction can be significantly improved with a GPU, but it needs Python 3.10+.
The environment upgrade is not easy. We have to modularize DeepCell. It is also interesting that we can swap this cell segmentation module with other models, such as cellpose.
At the beginning of the project, we based the entire pipeline on the DeepCell image, allowing us to investigate the implementation of DeepCell in greater detail. However, we are now trying to utilize a GPU to optimize the pipeline, and the DeepCell image has become a pain point due to the use of Python 3.8 by TensorFlow.
The UMAP construction can be significantly improved with a GPU, but it needs Python 3.10+.
The environment upgrade is not easy. We have to modularize DeepCell. It is also interesting that we can swap this cell segmentation module with other models, such as cellpose.