ICDAR2019-ArT-Recognition-Code for PKU Team Zero

Introduction

This is the code repository for our algorithm that ranked No.1 on ICDAR2019 Robust Reading Challenge on Arbitrary-Shaped Text (Latin scripts). Our team name is PKU Team Zero.

For more technical details, please refer to our paper: Rethinking Irregular Scene Text Recognition. We hope our efforts will act as a stepping stone to better recognition algorithms.

Team

• Shangbang Long (龙上邦), master student at MLD CMU.
• Yushuo Guan (关玉烁), master student at EECS, Peking University.
• Bingxuan Wang (王炳宣), junior undergraduate student at Yuanpei College, Peking University.
• Kaigui Bian (边凯归), associate professor at EECS, Peking University.
• Cong Yao (姚聪), algorithm team leader at MEGVII (Face++) Inc.

Competition Ranking

For full List, click here.

Method	Result	Total words	Correct words
PKU Team Zero (ours)	74.30%	35284	26216
CUTeOCR	73.91%	35284	26078
CRAFT (Preprocessing) + TPS-ResNet	73.87%	35284	26063
serial_rec	72.89%	35284	25717

Experiment Replication

Environment

Find the enclosed environment file, and use the following command to install:

conda env create -f environment.yml

You will need Anaconda to do so.

Data

In this section, we introduce how to prepare data for experiments. For download datasets and pretrained models, please refer to the Pretrained Models and Data section below.

All datasets should be placed under the ./dataset folder. The data should be arranged as follows:

- File Tree

|-dataset
|    |-dataset_name
|         | Label.json
|         |-IMG
|                1. jpg
|                2. jpg
|                ...

- Label.json File

The Structure of Label.json should be:

[
    {
        "img": "IMG/x.jpg",
        "word": str,
        "poly_x": [int, int, int, ...], 
        "poly_y": [int, int, int, ...],
        "chars": list(list(list(int)))
    }
]

The attributes are:

• img: path of the image file
• str: text content of the image
• poly_x: x coordinates of the bounding polygon, if exists.
• poly_y: y coordinates of the bounding polygon, if exists.
• chars: a 3-D 2x4xN array (list) representing bounding boxes of N characters. The 3 dimensions are: x/y coordinates, 4 corners, N characters, if exists.

How to Replicate the Experiments Presented in Our Paper:

You can refer to the paper for more details. To run experiments, find the training scripts under the corresponding folders, and call from the root folder, e.g. bash Experiment/Experiment1/Exper_1_CRNN_all_synth.sh.

Exp 1:

Experiments W.R.T. new synthetic datasets as described in Section 3.1.2-3.1.3 can be found in Experiment/Experiment1.

Exp 2:

Experiments W.R.T. mixing synthetic datasets and real world data as described in Section 3.2 can be found in Experiment/Experiment2.

Exp 3&4:

Experiments with model modifications as described in Section 4 can be found in Experiment/Experiment3 and Experiment/Experiment4.

ICDAR 2019-ArT:

To replicate our ICDAR 2019 models, readers can use scripts in Experiment/Experiment6.

Pretrained Models and Data

Pretrained Models

We select and release 3 representative models. Download from the link, unzip the file, and put the model files (ends with pth.tar) under th pretrained_models folder.

Models	Link
Rect trained with CurvedSynth + Synth90K Rect trained with CurvedSynth + Synth90K plus 15% real world data Rect with squarization and random rotation, trained on synthetic data	Google Drive

The use of pretrained models is demonstrated in the following scripts:

• ./Experiment/Experiment1/Exper_1_STN_all_synth_test.sh
• ./Experiment/Experiment2/Exper_2_STN_real_15_test.sh
• ./Experiment/Experiment4/Exper_4_Square_flip_test.sh

Data

We will release all datasets we used, for the convenience of the research community.

However, as they are large, we only release the following ones for now. We will update soon.

Dataset name	Description	Baidu Pan	Google Drive
RectTotal	Total-Text rectified by TextSnake	N/A	Google Drive
CurvedSynth(now full)	The newly proposed synthetic dataset we used	Baidu Drive password:9fp2	Part1 Part2 Part3

You can download and put these one under the dataset folder to start trying our code.

Curved SynthText Engine

As is discussed in detail in the paper, the Curved SynthText Engine we modified from the original SynthText is our trump card. We also opensource this engine at this repo.

Using synthetic images from this engine, we can expect 10+% improvement on Total-Text using a very simple algorithm.

RectTotal

We propose to evaluate algorithms on images rectified by TextSnake, as an investigation of key factors in text recognition. The TextSnake paper refers to the following one:

@inproceedings{long2018textsnake,
  title={Textsnake: A flexible representation for detecting text of arbitrary shapes},
  author={Long, Shangbang and Ruan, Jiaqiang and Zhang, Wenjie and He, Xin and Wu, Wenhao and Yao, Cong},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  pages={20--36},
  year={2018}
}

We release the Rectified Total-Text (RectTotal) dataset for further research.

Citation

If our paper and code help you in your research and understand the text recognition better, you are highly encouraged (though not required) to cite our paper:

@article{long2019ArT,
  title={Rethinking Irregular Scene Text Recognition},
  author={Long, Shangbang and Guan, Yushuo and Wang, Bingxuan and Bian, Kaigui and Yao, Cong},
  journal={arXiv preprint arXiv:1908.11834},
  year={2019}
}