CTW1500

# CTW1500 (MMOCR format) [CTW1500](https://github.com/Yuliang-Liu/Curve-Text-Detector) is a curve scene text detection benchmark with 1,000 training and 500 test images. The official download links (Box / CloudStor) are currently unreliable, so this repository re-hosts the dataset **already prepared in [MMOCR](https://github.com/open-mmlab/mmocr) format** for convenience. ## Contents `ctw1500_mmocr.zip` unpacks to: ```text ctw1500/ ├── textdet_imgs/ │ ├── train/ # 1000 images │ └── test/ # 500 images ├── textdet_train.json # MMOCR textdet training annotations └── textdet_test.json # MMOCR textdet test annotations ``` ## Usage ```bash pip install modelscope modelscope download --dataset WangXinhan/CTW1500 --local_dir ./ctw1500_dl unzip ctw1500_dl/ctw1500_mmocr.zip -d data/ ``` Then point MMOCR configs at `data/ctw1500` (the default `data_root='data/ctw1500'` in the textdet configs works out of the box). The same archive is mirrored on [HuggingFace](https://huggingface.co/datasets/HB16888/CTW1500). ## License The dataset is provided for academic research purposes, following the original CTW1500 release terms. The preparation pipeline is the one shipped with MMOCR (`tools/dataset_converters/prepare_dataset.py ctw1500 --task textdet`).