mask_rcnn_det

Mask-RCNN Detector

This code repository contains an implementation of Mask-RCNN based Text detector.

Preparing Dataset

Original images can be downloaded from: ICDAR2015, Total-Text , SCUT-CTW1500.

The formatted training datalist and test datalist can be found in demo/text_detection/datalist/

Training

Modified the paths ("imgs"/ "pretrained_model"/ "work_space", etc.) in the config files demo/text_detection/mask_rcnn_det/config/mask_rcnn_r50_fpn.py.

Run the following bash command in the command line,

cd $DAVAR_LAB_OCR_ROOT$/demo/text_detection/mask_rcnn_det/
bash dist_train.sh

We provide the implementation of online validation. If you want to close it to save training time, you may modify the startup script to add --no-validate command.

Offline Inference and Evaluation

We provide a demo of forward inference and visualization. You can modify the paths (test_dataset, image_prefix, etc.) in the testing script, and start testing:

python test.py 

Some visualization of detection results are shown:

Offline Evaluation

The offline evaluation tool can be found in davarocr/demo/text_detection/evaluation/.

Trained Model Download

All of the models are re-implemented and well trained in the based on the opensourced framework mmdetection.

Results on various datasets and trained models download:

Dataset	Backbone	Pretrained	Test Scale	Precision	Recall	Hmean	Links
Total-Text	ResNet-50	COCO	L-1350	84.90	82.26	83.56	config, pth (Access Code: 5fxU)
Total-Text	ResNet-101	COCO	L-1350	86.55	81.90	84.16	config, pth (Access Code: 9G1L)
SCUT-CTW1500	ResNet-50	COCO	L-1080	83.22	77.15	80.07	config, pth (Access Code: 2Qd8)
SCUT-CTW1500 (considers NOT CARE)	ResNet-50	COCO	L-1080		87.65	77.63	82.33
SCUT-CTW1500	ResNet-101	COCO	L-1080	82.76	78.7	80.68	config, pth (Access Code: 8wz2)
SCUT-CTW1500 (considers NOT CARE)	ResNet-101	COCO	L-1080		87.24	79.22	83.04
ICDAR 2015	ResNet-50	COCO	L-1600	89.19	77.03	82.67	config, pth (Access Code: 4R2B)
ICDAR 2015	ResNet-101	COCO	L-1600	90.43	75.11	82.06	config, pth (Access Code: 5H5z)

Pretrained model on COCO can be download in the repository of mmdetection/mask_rcnn.

Citation

@inproceedings{He_2017,
  title={Mask R-CNN},
  author={He, Kaiming and Gkioxari, Georgia and Dollar, Piotr and Girshick, Ross},
  booktitle={2017 IEEE International Conference on Computer Vision (ICCV)},
  year={2017}
}

License

This project is released under the Apache 2.0 license

Contact

If there is any suggestion and problem, please feel free to contact the author with [email protected].