chinese_text_recognition.md

Chinese-Text-Recognition

Data Downloading

Following the setup in Benchmarking-Chinese-Text-Recognition, we use the same training, validation and evaliation data as described in Datasets section.

Please download the following LMDB files as introduced in Downloads section:

• scene datasets: The union dataset contains RCTW, ReCTS, LSVT, ArT, CTW
• web: MTWI
• document: generated with Text Render
• handwriting dataset: SCUT-HCCDoc

Data Structure

After downloading the files, please put all training files under the same folder training, all validation data under validation folder, and all evaluation data under evaluation.

The data structure should be like:

chinese-text-recognition/
├── evaluation
│   ├── document_test
|   |   ├── data.mdb
|   │   └── lock.mdb
│   ├── handwriting_test
|   |   ├── data.mdb
|   │   └── lock.mdb
│   ├── scene_test
|   |   ├── data.mdb
|   │   └── lock.mdb
│   └── web_test
|       ├── data.mdb
|       └── lock.mdb
├── training
│   ├── document_train
|   |   ├── data.mdb
|   │   └── lock.mdb
│   ├── handwriting_train
|   |   ├── data.mdb
|   │   └── lock.mdb
│   ├── scene_train
|   |   ├── data.mdb
|   │   └── lock.mdb
│   └── web_train
|       ├── data.mdb
|       └── lock.mdb
└── validation
    ├── document_val
    |   ├── data.mdb
    │   └── lock.mdb
    ├── handwriting_val
    |   ├── data.mdb
    │   └── lock.mdb
    ├── scene_val
    |   ├── data.mdb
    │   └── lock.mdb
    └── web_val
        ├── data.mdb
        └── lock.mdb

Data Configuration

To use the datasets, you can specify the datasets as follow in configuration file.

Model Training

...
train:
  ...
  dataset:
    type: LMDBDataset
    dataset_root: dir/to/chinese-text-recognition/                    # Root dir of training dataset
    data_dir: training/                                               # Dir of training dataset, concatenated with `dataset_root` to be the complete dir of training dataset
...
eval:
  dataset:
    type: LMDBDataset
    dataset_root: dir/to/chinese-text-recognition/                    # Root dir of validation dataset
    data_dir: validation/                                             # Dir of validation dataset, concatenated with `dataset_root` to be the complete dir of validation dataset
  ...

Model Evaluation

...
train:
  # NO NEED TO CHANGE ANYTHING IN TRAIN SINCE IT IS NOT USED
...
eval:
  dataset:
    type: LMDBDataset
    dataset_root: dir/to/chinese-text-recognition/             # Root dir of evaluation dataset
    data_dir: evaluation/                                      # Dir of evaluation dataset, concatenated with `dataset_root` to be the complete dir of evaluation dataset
  ...

Back to dataset converters