wechat_jump_tensorflow

Deep Learning - WeChat Tiaotiao

Updated on 2018.01.14

simple.js has been added to the simple directory. The idea is the same as simple.py . It is written in JavaScript . Just install Auto.js on Android and run the script. The advantage is that it can be run directly on the mobile phone without being connected to a computer.

Updated on 2018.01.05

The annotated data was increased to 1200 images, and retrained with the more accurate faster_rcnn_inception_v2_coco model.

Project introduction

Zhihu article: https://zhuanlan.zhihu.com/p/32553763

Thanks to Chao and Fendudou for their contribution to labeling data.

Required environment

• Python3.6 , OpenCV2 , TensorFlow , etc.
• adb , used to debug Android phones, refer to https://github.com/wangshub/wechat_jump_game

Document introduction

simple.py in the simple directory uses OpenCV2 to detect the positions of chess pieces and target blocks. It is simple and crude. simple_ios.py is the corresponding IOS version.

The tensorflow directory includes the following files:

• wechat_jump_label_map.pbtxt : object category mapping file;
• utils : files that provide auxiliary functions;
• frozen_inference_graph_frcnn_inception_v2_coco.pb : trained object detection model, with a total of 1200 annotated data, trained using faster_rcnn_inception_v2_coco ;
• wechat_auto_jump.py : code to automatically jump

label.zip provides labeling tools. Use labelImg for object detection and labeling. For usage methods, please refer to物体检测标注说明.pdf