Unsupervised human HT29 colon cancer cell morphology classification based on Deep InfoMax (link to paper)
Tested on Ubuntu 20.04 with pyTorch 1.12 and CUDA 11.6
2023-04-27 update:
- Improved code readability.
2023-04-25 update:
- Added prior distribution learning;
- Added binary GMM for data preparation;
- Added UMAP for vector dimension reduction;
- Added N-class GMM for morphology clustering;
- Modified dataloader for loading 4D batch;
- Modified network structures of encoder and summarizer;
- Modified loss functions accordingly (see report).
- Update README and Makefile.
- Navigate to the repo root;
- Copy all images in the dataset to ./datasets/Images;
- In the terminal, run "make";
- You should see a new tag in browser pops up.
- Run "prepare.ipynb" to generate binary masks via cell →run all;
- Run "train.ipynb" to train all models via cell →run all;
- Wait until the training completes, use visdom to check status.
- Run "test.ipynb" via cell →run all;
- Wait until the testing completes, results can be found under ./saves.