ECE271B project
PyTorch implementation for self-supervised part segmentation via keypoint constraints - Application to histopathology study in cancer detection.
The code is developed based on Pytorch v1.8+ with TensorboardX as visualization tools. We recommend to use conda env to run the code:
$ conda env create python3 -n spask_env
$ source spask_env/bin/activate
(spask_env)$ pip install -r requirements.txt
To deactivate the virtual environment, run $conda deactivate
. To activate the environment again, run $ conda activate spask_env
.
$ ./download_CelebA.sh
Download CelebA unaligned from here.
$ CUDA_VISIBLE_DEVICES={GPU} python train.py -f exps/SPASK_K8_train_geometric.json
where {GPU}
is the GPU device number.
The code for experiments on Camelyon16 dataset is in cam16 folder
Code is largely based on SCOPS paper
Apache 2.0 license