- Problem: The goal is to create an algorithm to identify metastatic cancer in small image patches taken from larger digital pathology scan.
- Dataset: The PatchCamelyon benchmark (Pcam), derived from the Camelyon16 Challenge, has following features:
- type: HDF5 File
- input: 96x96
- train/valid/test: 262144 / 32768 / 32768
- label: binary (positive 1 / negative 0)
- Our Approach：
- Using transfer learning with ResNet-50 as our backbone (pretrained by imagenet)
- Discriminative learning rate to finetune
- Using Adam as optimizer and Cross Entropy as our loss function
- Using Learning rate decay to achieve better convergence
- Replace ReLU() activation with SELU() in fully connected layer.
- To alleviate the over-fitting problem, we use Data Augmentation ( > 10), Weight Decay, Dropout.
# recommand using anaconda to create virtual environment
conda create -n pcam python=3.7
conda activate pcam
# install package through tsinghua channel
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
# run train/test, single-gpu version
python train.py
python test.py
basveeling/pcam