By Fanxin Xu, Weixuan Wu, Beibei Liu, He Lyu and Wei Xiang
This repository is an official implementation of the paper MJOD-2136: A New Dataset and A light-weight model for mahjong object detection.
MJOD-2136: a new mahjong object detection dataset with COCO format for research purpose.
MJOD-Net: a new light weight model based on YOLOF architecture implemented by MMDetection for mahjong object detection.
-
read datasets/README.md for more details about our datasets: MJOD-2136.
-
see INSTALL.md for the preparation of environment based on MMDetection.
-
Before training, we should download datasets: MJOD-2136 and prepare the config files.
datasets(Google Drive): https://drive.google.com/drive/folders/1kfLVlEjWWPz9SijYhO-M9zHNV4n8Xema?usp=sharing
datasets(Baidu Disk): https://pan.baidu.com/s/1TAihGvfxj-jwwQl0qaix9g - co4p
pth(Baidu Disk): https://pan.baidu.com/s/1PNt5PqzM_yx3pNO8yIMBHQ - uzqn
Before training and testing, the MMDetection environment should be implemented while the above files in folders should load in the corresponding folders.(e.g. /configs/MJOD_Net -> mmdetection/configs/)
-
To train the MJOD-Net, run following command
python tools/train.py ${CONFIG_FILE}
-
To test the MJOD-Net, run following command
python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE}
This project is released under the Apache 2.0 license.