Use Custom Datasets

Datasets that have builtin support in X-modaler are listed in builtin datasets. If you want to use a custom dataset while also reusing X-modaler’s data loaders, you will need to register your dataset (i.e., tell X-modaler how to obtain your dataset).

Register a Dataset

To let X-modaler know how to obtain a dataset named “MyDataset”, users need to implement a class either in the folder xmodaler/datasets/images or xmodaler/datasets/videos that returns the item in your dataset and then tell X-modaler about this class:

from ..build import DATASETS_REGISTRY

__all__ = ["MyDataset"]

@DATASETS_REGISTRY.register()
class MyDataset:
    @configurable
    def __init__(
        self,
        stage: str,
        anno_file: str,
        seq_per_img: int,
        max_feat_num: int,
        max_seq_len: int,
        feats_folder: str
    ):
        # load dataset and set hyperparameters
        ...
	
    @classmethod
    def from_config(cls, cfg, stage: str = "train"):
        ...
        return arguments

    def __call__(self, dataset_dict):
        # process a data item for a specific task
        ...
        return item_dict

The function can do arbitrary things and should return the data dict in either of the following formats:

• X-modaler’s standard dataset dict using builtin keys defined by xmodaler.config.kfg. This will make it work with many other builtin features in X-modaler, so it’s recommended to use it when it’s sufficient.

• Any custom format. You can also return arbitrary dict in your own format by adding extra keys to xmodaler.config.kfg for new tasks. Then you will need to handle them properly downstream as well.