Feature Pyramid Networks for Pulmonary Nodules Detection

Configuration Environment

this project can be trained on Google Colab,while you should turn your VPN on.

Installation

Clone the repository

git clone https://github.com/Andyato/Pulmonary-Nodule-Detection

Make tfrecord

The data is VOC format

├── VOCdevkit
│   ├── VOCdevkit_train
│       ├── Annotation
│       ├── JPEGImages
│    ├── VOCdevkit_test
│       ├── Annotation
│       ├── JPEGImages

cd $ROOT/data/io/  
python convert_data_to_tfrecord.py --VOC_dir='***/VOCdevkit/VOCdevkit_train/' --save_name='train' --img_format='.jpg' --dataset='nodules'

Train

1、Modify $ROOT/libs/lable_name_dict/***_dict.py, corresponding to the number of categories in the configuration file
2、download pretrain weight(resnet_v1_101_2016_08_28.tar.gz or resnet_v1_50_2016_08_28.tar.gz) from here, then extract to folder $ROOT/data/pretrained_weights
3、

cd $ROOT/tools
python train.py

Test tfrecord

cd $ROOT/tools    
python test.py

eval

cd $FPN_ROOT/tools   
python eval.py --weights (your trained weights directory path) --image_num (eval images number)

Summary

tensorboard --logdir=$ROOT/output/res**_summary/

Test results

Reference

1、tf-faster-rcnn 2、FastMaskRCNN 3、object_detection

Andyato / Pulmonary-Nodule-Detection