Skip to content

[CVPR 2021] Rethinking Text Segmentation: A Novel Dataset and A Text-Specific Refinement Approach

Notifications You must be signed in to change notification settings

SHI-Labs/Rethinking-Text-Segmentation

Repository files navigation

Rethinking Text Segmentation: A Novel Dataset and A Text-Specific Refinement Approach

This is the repo to host the dataset TextSeg and code for TexRNet from the following paper:

Xingqian Xu, Zhifei Zhang, Zhaowen Wang, Brian Price, Zhonghao Wang and Humphrey Shi, Rethinking Text Segmentation: A Novel Dataset and A Text-Specific Refinement Approach, ArXiv Link

Note:

[2021.04.21] So far, our dataset is partially released with images and semantic labels. Since many people may request the dataset for OCR or non-segmentation tasks, please stay tuned, and we will release the dataset in full ASAP.

[2021.06.18] Our dataset is now fully released. To download the data, please send a request email to textseg.dataset@gmail.com and tell us which school you are affiliated with. Please be aware the released dataset is version 2, and the annotations are slightly different from the one in the paper. In order to provide the most accurate dataset, we went through a second round of quality assurance, in which we fixed some faulty annotations and made them more consistent across the dataset. Since our TexRNet in the paper doesn't use OCR and character instance labels (i.e. word- and character-level bounding polygons; character-level masks;), we will not release the older version of these labels. However, we release the retroactive semantic_label_v1.tar.gz for researchers to reproduce the results in the paper. For more details about the dataset, please see below.

Introduction

Text in the real world is extremely diverse, yet current text dataset does not reflect such diversity very well. To bridge this gap, we proposed TextSeg, a large-scale fine-annotated and multi-purpose text dataset, collecting scene and design text with six types of annotations: word- and character-wise bounding polygons, masks and transcriptions. We also introduce Text Refinement Network (TexRNet), a novel text segmentation approach that adapts to the unique properties of text, e.g. non-convex boundary, diverse texture, etc., which often impose burdens on traditional segmentation models. TexRNet refines results from common segmentation approach via key features pooling and attention, so that wrong-activated text regions can be adjusted. We also introduce trimap and discriminator losses that show significant improvement on text segmentation.

TextSeg Dataset

Image Collection

Annotation

Download

Our dataset (TextSeg) is academia-only and cannot be used on any commercial project and research. To download the data, please send a request email to textseg.dataset@gmail.com and tell us which school you are affiliated with.

A full download should contain these files:

  • image.tar.gz contains 4024 images.
  • annotation.tar.gz labels corresponding to the images. These three types of files are included:
    • [dataID]_anno.json contains all word- and character-level translations and bounding polygons.
    • [dataID]_mask.png contains all character masks. Character mask label value will be ordered from 1 to n. Label value 0 means background, 255 means ignore.
    • [dataID]_maskeff.png contains all character masks with effect.
    • Adobe_Research_License_TextSeg.txt license file.
  • semantic_label.tar.gz contains all word-level (semantic-level) masks. It contains:
    • [dataID]_maskfg.png 0 means background, 100 means word, 200 means word-effect, 255 means ignore. (The [dataID]_maskfg.png can also be generated using [dataID]_mask.png and [dataID]_maskeff.png)
  • split.json the official split of train, val and test.
  • [Optional] semantic_label_v1.tar.gz the old version of label that was used in our paper. One can download it to reproduce our paper results.

TexRNet Structure and Results

In this table, we report the performance of our TexRNet on 5 text segmentation dataset including ours.

TextSeg(Ours) ICDAR13 FST COCO_TS MLT_S Total-Text
Method fgIoUF-score fgIoUF-score fgIoUF-score fgIoUF-score fgIoUF-score
DeeplabV3+ 84.070.914 69.270.802 72.070.641 84.630.837 74.440.824
HRNetV2-W48 85.030.914 70.980.822 68.930.629 83.260.836 75.290.825
HRNetV2-W48 + OCR 85.980.918 72.450.830 69.540.627 83.490.838 76.230.832
Ours: TexRNet + DeeplabV3+ 86.06 0.921 72.16 0.835 73.980.722 86.31 0.830 76.53 0.844
Ours: TexRNet + HRNetV2-W48 86.840.924 73.380.850 72.39 0.720 86.09 0.865 78.470.848

To run the code

Set up the environment

conda create -n texrnet python=3.7
conda activate texrnet
pip install -r requirement.txt

To eval

First, make the following directories to hold pre-trained models, dataset, and running logs:

mkdir ./pretrained
mkdir ./data
mkdir ./log

Second, download the models from this link. Move those downloaded models to ./pretrained.

Thrid, make sure that ./data contains the data. A sample root directory for TextSeg would be ./data/TextSeg.

Lastly, evaluate the model and compute fgIoU/F-score with the following command:

python main.py --eval --pth [model path] [--hrnet] [--gpu 0 1 ...] --dsname [dataset name]

Here is the sample command to eval a TexRNet_HRNet on TextSeg with 4 GPUs:

python main.py --eval --pth pretrained/texrnet_hrnet.pth --hrnet --gpu 0 1 2 3 --dsname textseg

The program will store results and execution log in ./log/eval.

To train

Similarly, these directories need to be created:

mkdir ./pretrained
mkdir ./pretrained/init
mkdir ./data
mkdir ./log

Second, we use multiple pre-trained models for training. Download these initial models from this link. Move those models to ./pretrained/init. Also, make sure that ./data contains the data.

Lastly, execute the training code with the following command:

python main.py [--hrnet] [--gpu 0 1 ...] --dsname [dataset name] [--trainwithcls]

Here is the sample command to train a TexRNet_HRNet on TextSeg with classifier and discriminate loss using 4 GPUs:

python main.py --hrnet --gpu 0 1 2 3 --dsname textseg --trainwithcls

The training configs, logs, and models will be stored in ./log/texrnet_[dsname]/[exid]_[signature].

Bibtex

@article{xu2020rethinking,
  title={Rethinking Text Segmentation: A Novel Dataset and A Text-Specific Refinement Approach},
  author={Xu, Xingqian and Zhang, Zhifei and Wang, Zhaowen and Price, Brian and Wang, Zhonghao and Shi, Humphrey},
  journal={arXiv preprint arXiv:2011.14021},
  year={2020}
}

Acknowledgements

The directory .\hrnet_code is directly copied from the HRNet official github website (link). HRNet code ownership should be credited to HRNet authors, and users should follow their terms of usage.

About

[CVPR 2021] Rethinking Text Segmentation: A Novel Dataset and A Text-Specific Refinement Approach

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published