Submodule of Dataset

Modules of dataset, inherit from torch.utils.data.Dataset, which is an iteration class provides batches of data to model.

A custom dataset module can be also easily used by adding two lines of code:

from dolphin.utils import Registers

@Registers.dataset.register
class CustomDataset():

    def __init__(self):
        ...

Dataflow Pipeline

There are more than 10 methods of data preprocessing methods implemented here, such as image resizing, random image flip, random cropping and so on, which are design with inheritance from BaseModule. All of them take a python dictionary as input and output with the same type. If customization of pipeline module is needed, the new method must go by the same shape as said before.