1. We highly recommend to use docker image because our model contains custom operation which depends on framework and cuda version.
2. We provide trained model for ICDAR 2017, 2013 which is in final_checkpoint_ch8 and for ICDAR 2015 which is in final_checkpoint_ch4
3. This code is mainly focused on inference. To train our model, training gpu like V100 is needed. please check our paper in detail.

1. Nvidia-docker
2. Tensorflow 1.14
3. Miminum GPU requirement : NVIDIA GTX 1080TI

• Make docker image and container

docker build --tag rbimage ./dockerfile
docker run --runtime=nvidia --name rbcontainer -v /rotated-box-is-back-path:/rotated-box-is-back -i -t rbimage /bin/bash

• build custom operations in container

cd /rotated-box-is-back/nms 
cmake ./
make
./shell.sh

cd /rotated-box-is-back/
python viz.py --test_data_path=./sample --checkpoint_path=./final_checkpoint_ch8 --output_dir=./sample_result  --thres 0.6 --min_size=1600 --max_size=2000

1. please replace icdar_testset_path to your-icdar-2017-testset-folder path.

python viz.py --test_data_path=icdar_testset_path --checkpoint_path=./final_checkpoint_ch8 --output_dir=./ic17  --thres 0.6 --min_size=1600 --max_size=2000

1. please replace icdar_testset_path to your-icdar-2015-testset-folder path.
2. To converting evalutation format. Convert result text file like below

python viz.py --test_data_path=icdar_testset_path --checkpoint_path=./final_checkpoint_ch4 --output_dir=./ic15  --thres 0.7 --min_size=1100 --max_size=2000
python text_postprocessing.py -i=./ic15/ -o=./ic15_format/ -e True

1. please replace icdar_testset_path to your-icdar-2013-testset-folder path.
2. To converting evalutation format. Convert result text file like below

python viz.py --test_data_path=icdar_testset_path --checkpoint_path=./final_checkpoint_ch8 --output_dir=./ic13  --thres 0.55 --min_size=700 --max_size=900
python text_postprocessing.py -i=./ic13/ -o=./ic13_format/ -e True -m rec

1. It can be trained below command line

python train_refine_estimator.py --input_size=1024 --batch_size=2 --checkpoint_path=./finetuning --training_data_path=your-image-path --training_gt_path=your-gt-path  --learning_rate=0.00001 --max_epochs=500  --save_summary_steps=1000 --warmup_path=./final_checkpoint_ch8

This work was supported by Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No. 1711125972, Audio-Visual Perception for Autonomous Rescue Drones).

If you found it is helpfull for your research, please cite:

Lee J., Lee J., Yang C., Lee Y., Lee J. (2021) Rotated Box Is Back: An Accurate Box Proposal Network for Scene Text Detection. In: Lladós J., Lopresti D., Uchida S. (eds) Document Analysis and Recognition – ICDAR 2021. ICDAR 2021. Lecture Notes in Computer Science, vol 12824. Springer, Cham. https://doi.org/10.1007/978-3-030-86337-1_4