Dataset: ReCTS2019

Tasks - ICDAR 2019 Robust Reading Challenge on Reading Chinese Text on Signboard

Reference

• example scripts

The structure of data_processing folder as below.

.
├── README.md
├── change_json.py
├── classes.txt
├── dictionary_inv.json
├── images
│   ├── train_ReCTS_000001.jpg
│   ├── train_ReCTS_000002.jpg
│   ├── train_ReCTS_000003.jpg
│   ├── train_ReCTS_000004.jpg
│   └── train_ReCTS_000005.jpg
├── labels
│   ├── train_ReCTS_000001.json
│   ├── train_ReCTS_000002.json
│   ├── train_ReCTS_000003.json
│   ├── train_ReCTS_000004.json
│   └── train_ReCTS_000005.json
├── my_Bezier_generator2.py
├── my_generate_abcnet_json.py
└── pre_processing.py

Step one: Preprocessing

python pre_processing.py

Step two: Processing the custom dataset with only four vertices

python change_json.py

Step three: Given polygonal annotation, generating bezier curve annotation

python my_Bezier_generator2.py

Step four: Given bezier curve annotation, generating coco-like annotation format for training abcnet

python my_generate_abcnet_json.py ./ train 0

python my_generate_abcnet_json.py ./ train 
python my_generate_abcnet_json.py ./ test

Step five: Rename

python rename.py

Deal with gt

find ./gt_rects/ -name '*' -print | zip -j gt_rects.zip -@