/ Machine Learning

Arbitrary-Oriented Scene Text Detection via Rotation Proposals

Arbitrary-Oriented Scene Text Detection via Rotation Proposals

RRPN in pytorch

RRPN in pytorch, which is implemented into facebook's benchmark: https://github.com/facebookresearch/maskrcnn-benchmark.


  • From original repo: In pytorch 1.0, Somehow faster than original repo in both training and inference.
  • Training and evaluation checked: Testing in IC15 with training data in {IC13, IC15, IC17mlt}, and receives Fscore of 83% vs. 81% in caffe repo.
  • What's new: RRoI Pooling is replaced with RRoI Align(bilinear interpolation for sampling), FPN structure supported, easy to change various backbones for different purposes.

Configuring your dataset

  • Your dataset path can be set in $RRPN_ROOT/maskrcnn_benchmark/config/paths_catalog.py. We implemented interface for {IC13, IC15, IC17mlt, LSVT, ArT} for common use(Start from line 96):
 "RRPN_train": {  # including IC13 and IC15
                # 'IC13': 'Your dataset path',
            "split": 'train'
  • Add your dataset?
    You need to form a dict array as follows:
im_info = {
    'gt_classes': your class_id array,
    'max_classes': your class_id array,
    'image': path to access one image,
    'boxes': rotate box in {cx, cy, w, h, θ},
    'flipped': Not supported, just False, 
    'gt_overlaps': overlaps fill with 1 (gt with gt),
    'seg_areas': H * W for an rbox,
    'height': height of an image,
    'width': width of an image,
    'max_overlaps': overlaps fill with 1 (gt with gt),
    'rotated': just True

Examples can be seen in $RRPN_ROOT/maskrcnn_benchmark/data/rotation_series.py
Your data API should be add to the variable DATASET:

    'Your Dataset Name': 'Your Dataset API'


# In your root of RRPN
python tools/train_net.py --config-file=configs/rrpn/e2e_rrpn_R_50_C4_1x_ICDAR13_15_17_trial.yaml
  • Multi-GPU phase is not testing yet, be careful to use GPU more than 1.


  • Using $RRPN_ROOT/demo/RRPN_Demo.py to test images you want. The demo will generate a text for your detected coodinates.
  • Showing the detected image by ture the variable vis to True.


  • Enjoy it with all the codes.
  • citing us if you find it work in your projects.
    author = {Jianqi Ma},
    title = {{RRPN in pytorch}},
    year = {2019},
    howpublished = {\url{https://github.com/mjq11302010044/RRPN_pytorch}},
    Author = {Jianqi Ma and Weiyuan Shao and Hao Ye and Li Wang and Hong Wang and Yingbin Zheng and Xiangyang Xue},
    Title = {Arbitrary-Oriented Scene Text Detection via Rotation Proposals},
    journal = {IEEE Transactions on Multimedia},