An Implentation of RetinaNet in Pytorch

optional backbone:

• se_resnext50_32x4d
• se_resnext101_32x4d

usage:

setup.sh: compile cython code
config.json: config file
train.py: main train script
dataGen/data_loader.py: Subclass torch.utils.data.Dataset or modify class RsnaDataset for your owen dataset.
prepare_data.ipynb: data processing script for RSNA Pneumonia Detection Challenge
demo.ipynb: sample code showing how to predict with model.predict method

---

Application

train this model with dataset of RSNA Pneumonia Detection Challenge

credits:

1. Cadene pretrained-models.pytorch
2. fizyr/keras-retinanet
3. Squeeze-and-Excitation Networks
4. Focal Loss for Dense Object Detection