This is a simple python demo for Agora 1-to-1 video call extended by Tensorflow object detection API.

其他语言版本： 简体中文

• Python 3.6
• Tensorflow >= 1.12
• opencv-python
• pillow

• Get Agora Python SDK from here and put the corresponding files to the root directory of this project. For Windows, you need .pyd and .dll files. For Mac, you need the .so file.
• Download Tensorflow models and put the object_detection directory to the root directory of this project.
• Install Protobuf. Then run:

protoc object_detection/protos/*.proto --python_out=.

• Download pre-trained models from here. We recommend ssd_mobilenet_v1_coco and ssdlite_mobilenet_v2_coco since they are much faster.
• To extract the frozen graph, run:

python extractGraph.py --model_file='FILE_NAME_OF_YOUR_MODEL'

• Finally, specify your model name in the beginning of callBack.py and your AppId in the beginning of demo.py.