Testing different CapsNet implementation using patch_camelyon dataset from tensorflow_dataset.
Tested implementation:
- Implementation of "Matrix Capsules with EM Routing" by Ashley Gritzman.
- Efficient-CapsNet by Vittorio Mazzia and Francesco Salvetti. They implemented the original CapsNet from Hinton and a new version that exploit self-attention as routing mechanism.
It consists of 327.680 RGB color images (96 x 96px) extracted from histopathologic scans of lymph node sections. Each image is annoted with a binary label indicating presence of metastatic tissue. It comes alredy splitted in training, validation and test set.
Tensorflow_dataset provide a very interesting playground to visualize and interact with the dataset.
- Clone the git repository as ZIP or with the command line:
git clone https://github.com/AlbezJelt/AML2021_Capsule_Network
- Install the required packages:
pip install tensorflow tensorflow-datasets
- Run the dedicated download script:
cd AML2021_Capsule_Network python3 /data/download_dataset.py
Provided models require different versions of the same packages in order to run. Please refer to READMEs in dedicated folders (Efficient CapsNet and EM Routing).