# Writing Custom Dataset Importers This recipe demonstrates how to write a [custom DatasetImporter](https://voxel51.com/docs/fiftyone/user_guide/import_datasets.html#custom-dataset-importer) and use it to load a dataset from disk in your custom format into FiftyOne. ## Setup If you haven’t already, install FiftyOne: In this recipe we’ll use the [FiftyOne Dataset Zoo](https://docs.voxel51.com/dataset_zoo/index.html) to download the [CIFAR-10 dataset](https://www.cs.toronto.edu/~kriz/cifar.html) to use as sample data to feed our custom importer. Behind the scenes, FiftyOne either uses the [TensorFlow Datasets](https://www.tensorflow.org/datasets) or [TorchVision Datasets](https://pytorch.org/vision/stable/datasets.html) libraries to wrangle the datasets, depending on which ML library you have installed. You can, for example, install PyTorch as follows: ## Writing a DatasetImporter FiftyOne provides a [DatasetImporter](https://voxel51.com/docs/fiftyone/api/fiftyone.utils.data.html#fiftyone.utils.data.importers.DatasetImporter) interface that defines how it imports datasets from disk when methods such as [Dataset.from_importer()](https://voxel51.com/docs/fiftyone/api/fiftyone.core.html#fiftyone.core.dataset.Dataset.from_importer) are used. `DatasetImporter` itself is an abstract interface; the concrete interface that you should implement is determined by the type of dataset that you are importing. See [writing a custom DatasetImporter](https://voxel51.com/docs/fiftyone/user_guide/import_datasets.html#custom-dataset-importer) for full details. In this recipe, we’ll write a custom [LabeledImageDatasetImporter](https://voxel51.com/docs/fiftyone/api/fiftyone.utils.data.html#fiftyone.utils.data.importers.LabeledImageDatasetImporter) that can import an image classification dataset whose image metadata and labels are stored in a `labels.csv` file in the dataset directory with the following format: ```none filepath,size_bytes,mime_type,width,height,num_channels,label ,,,,,, ,,,,,, ... ``` Here’s the complete definition of the `DatasetImporter`: ## Generating a sample dataset In order to use `CSVImageClassificationDatasetImporter`, we need to generate a sample dataset in the required format. Let’s first write a small utility to populate a `labels.csv` file in the required format. Now let’s populate a directory with a `labels.csv` file in the format required by `CSVImageClassificationDatasetImporter` with some samples from the test split of CIFAR-10: Let’s inspect the contents of the labels CSV to ensure they’re in the correct format: ## Importing a dataset With our dataset and `DatasetImporter` in-hand, loading the data as a FiftyOne dataset is as simple as follows: ## Cleanup You can cleanup the files generated by this recipe by running: