README.md

Prepare datasets

It is recommended to symlink the dataset root to $PAN_PP.PYTORCH/data. If your folder structure is different, you may need to change the corresponding paths in dataloader files.

pan_pp.pytorch
└── data
    ├── CTW1500
    │   ├── train
    │   │   ├── text_image
    │   │   └── text_label_curve
    │   └── test
    │       ├── text_image
    │       └── text_label_curve
    ├── total_text
    │   ├── Images
    │   │   ├── Train
    │   │   └── Test
    │   └── Groundtruth
    │       ├── Polygon
    │       └── Rectangular
    ├── ICDAR2015
    │   └── Challenge4
    │       ├── ch4_training_images
    │       ├── ch4_training_localization_transcription_gt
    │       ├── ch4_test_images
    │       └── ch4_test_localization_transcription_gt
    ├── MSRA-TD500
    │   ├── train
    │   └── test
    ├── HUST-TR400
    ├── COCO-Text
    │   └── train2014
    ├── SynthText
    │   ├── 1
    │   ├── ...
    │   └── 200
    └── ICDAR2017-MLT
        ├── ch8_training_images
        ├── ch8_validation_images
        ├── ch8_training_localization_transcription_gt_v2
        └── ch8_validation_localization_transcription_gt_v2

Download

These datasets can be downloaded from the following links:

• CTW1500 [dataset]

• Total-Text [image] [gt]

• ICDAR2015 [dataset]

• MSRA-TD500 [dataset]

• HUST-TR400 [dataset]

• SynthText [dataset]

• ICDAR2017-MLT [dataset]

• COCO-Text [dataset]