Framework
We design a new framework to support LFD implementation. In this page, we make an overall description of our framework and present some key design ideas. This implementation obsorbs some excellent ideas from mmdetection.
In the beginning, the workflow is given below:
As we can see, the framework is raughly divided into three parts: data pipeline, traning and deploy.
Data pipeline provides functionalities for packing dataset, samplers, augmentation and data loader.
Training part is mainly about configuring settings, especially model setting and training setting.
Deploy supports onnx conversion and third-part Lib inference.
For more details, check the related sub pages.