R2CNN: Rotational Region CNN for Orientation Robust Scene Detection

We modified the R2CNN to make its performance more robust and excellent, and achieved better performance in the DOTA and icdar dataset. The code will be updated in https://github.com/DetectionTeamUCAS. Stay tuned!

A Tensorflow implementation of FPN or R2CNN detection framework based on FPN.
You can refer to the papers R2CNN Rotational Region CNN for Orientation Robust Scene Text Detection or Feature Pyramid Networks for Object Detection
Other rotation detection method reference R-DFPN, RRPN and R2CNN_HEAD
If useful to you, please star to support my work. Thanks.

Citing R-DFPN

If you find R-DFPN useful in your research, please consider citing:

@article{yangxue_r-dfpn:http://www.mdpi.com/2072-4292/10/1/132
    Author = {Xue Yang, Hao Sun, Kun Fu, Jirui Yang, Xian Sun, Menglong Yan and Zhi Guo},
    Title = {{R-DFPN}: Automatic Ship Detection in Remote Sensing Images from Google Earth of Complex Scenes Based on Multiscale Rotation Dense Feature Pyramid Networks},
    Journal = {Published in remote sensing},
    Year = {2018}
}  

Configuration Environment

ubuntu(Encoding problems may occur on windows) + python2 + tensorflow1.2 + cv2 + cuda8.0 + GeForce GTX 1080
If you want to use cpu, you need to modify the parameters of NMS and IOU functions use_gpu = False in cfgs.py
You can also use docker environment, command: docker pull yangxue2docker/tensorflow3_gpu_cv2_sshd:v1.0

Installation

Clone the repository

git clone https://github.com/yangxue0827/R2CNN_FPN_Tensorflow.git    

Make tfrecord

The image name is best in English.
The data is VOC format, reference here
data path format ($R2CNN_ROOT/data/io/divide_data.py)
VOCdevkit

VOCdevkit_train

Annotation
JPEGImages

VOCdevkit_test

Annotation
JPEGImages

Clone the repository

cd $R2CNN_ROOT/data/io/  
python convert_data_to_tfrecord.py --VOC_dir='***/VOCdevkit/VOCdevkit_train/' --save_name='train' --img_format='.jpg' --dataset='ship'
     

Demo

1、Unzip the weight $R2CNN_ROOT/output/res101_trained_weights/*.rar
2、put images in $R2CNN_ROOT/tools/inference_image
3、Configure parameters in $R2CNN_ROOT/libs/configs/cfgs.py and modify the project's root directory
4、

cd $R2CNN_ROOT/tools      

5、image slice

python inference1.py   

6、large image

cd $FPN_ROOT/tools
python demo1.py --src_folder=.\demo_src --des_folder=.\demo_des         

Train

1、Modify $R2CNN_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 $R2CNN_ROOT/data/pretrained_weights
3、

cd $R2CNN_ROOT/tools      

4、Choose a model(FPN or R2CNN)
If you want to train FPN :

python train.py   

elif you want to train R2CNN:

python train1.py   

Test tfrecord

cd $R2CNN_ROOT/tools   
python test.py(test1.py)   

eval

cd $R2CNN_ROOT/tools   
python eval.py(eval1.py)  

Summary

tensorboard --logdir=$R2CNN_ROOT/output/res101_summary/ 

Graph

icdar2015 test results

Test results