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
These datasets can be downloaded from the following links: