Waffle Hub

Waffle Hub provide two key component classes: Hub and Dataset.

Hub

Hub provides same interface for various powerfull Deep Learning Frameworks. Here is our brief system architecture.

Each input and output adapter is responsible for converting our interface to the framework's interface. For example, Ultralytics uses imgsz for image size parameter, but detectron2 uses IMAGE_SIZE. So, we need to convert our interface to the framework's interface. waffle_hub provides InputAdapter and OutputAdapter for this purpose.

Dataset

Dataset class support many types of data format such as coco, yolo. You can use it to convert dataset or manage dataset.