fiftyone.zoo.datasets.base.BDD100KDataset

class documentation

class BDD100KDataset(FiftyOneDataset): (source)

Constructor: BDD100KDataset(source_dir, copy_files)

The Berkeley Deep Drive (BDD) dataset is one of the largest and most diverse video datasets for autonomous vehicles.

The BDD100K dataset contains 100,000 video clips collected from more than 50,000 rides covering New York, San Francisco Bay Area, and other regions. The dataset contains diverse scene types such as city streets, residential areas, and highways. Furthermore, the videos were recorded in diverse weather conditions at different times of the day.

The videos are split into training (70K), validation (10K) and testing (20K) sets. Each video is 40 seconds long with 720p resolution and a frame rate of 30fps. The frame at the 10th second of each video is annotated for image classification, detection, and segmentation tasks.

This version of the dataset contains only the 100K images extracted from the videos as described above, together with the image classification, detection, and segmentation labels.

In order to load the BDD100K dataset, you must download the source data manually. The directory should be organized in the following format:

source_dir/
    labels/
        bdd100k_labels_images_train.json
        bdd100k_labels_images_val.json
    images/
        100k/
            train/
            test/
            val/

You can register at https://bdd-data.berkeley.edu in order to get links to download the data.

Example usage:

import fiftyone as fo
import fiftyone.zoo as foz

# The path to the source files that you manually downloaded
source_dir = "/path/to/dir-with-bdd100k-files"

dataset = foz.load_zoo_dataset(
    "bdd100k",
    split="validation",
    source_dir=source_dir,
)

session = fo.launch_app(dataset)

Dataset size: 7.10 GB
Source: https://bdd-data.berkeley.edu

Parameters
source_dir	the directory containing the manually downloaded BDD100K files
copy_files	whether to move (False) or create copies (True) of the source files when populating the dataset directory

Method	`__init__`	Undocumented
Instance Variable	`copy_files`	Undocumented
Instance Variable	`source_dir`	Undocumented
Property	`name`	The name of the dataset.
Property	`requires_manual_download`	Whether this dataset requires some files to be manually downloaded by the user before the dataset can be loaded.
Property	`supported_splits`	A tuple of supported splits for the dataset, or None if the dataset does not have splits.
Property	`tags`	A tuple of tags for the dataset.
Method	`_download_and_prepare`	Internal implementation of downloading the dataset and preparing it for use in the given directory.

Inherited from ZooDataset (via FiftyOneDataset):

Static Method	`get_info_path`	Returns the path to the `ZooDatasetInfo` for the dataset.
Static Method	`has_info`	Determines whether the directory contains `ZooDatasetInfo`.
Static Method	`load_info`	Loads the `ZooDatasetInfo` from the given dataset directory.
Method	`download_and_prepare`	Downloads the dataset and prepares it for use.
Method	`get_split_dir`	Returns the directory for the given split of the dataset.
Method	`has_split`	Whether the dataset has the given split.
Method	`has_tag`	Whether the dataset has the given tag.
Property	`has_patches`	Whether the dataset has patches that may need to be applied to already downloaded files.
Property	`has_splits`	Whether the dataset has splits.
Property	`has_tags`	Whether the dataset has tags.
Property	`importer_kwargs`	A dict of default kwargs to pass to this dataset's `fiftyone.utils.data.importers.DatasetImporter`.
Property	`is_remote`	Whether the dataset is remotely-sourced.
Property	`parameters`	An optional dict of parameters describing the configuration of the zoo dataset when it was downloaded.
Property	`supports_partial_downloads`	Whether the dataset supports downloading partial subsets of its splits.
Method	`_get_splits_to_download`	Undocumented
Method	`_is_dataset_ready`	Undocumented
Method	`_is_split_ready`	Undocumented
Method	`_patch_if_necessary`	Internal method called when an already downloaded dataset may need to be patched.

def __init__(self, source_dir=None, copy_files=True): (source) ¶

Undocumented

copy_files: True = (source) ¶

Undocumented

source_dir: None = (source) ¶

Undocumented

@property
name = (source) ¶

overrides fiftyone.zoo.datasets.ZooDataset.name

The name of the dataset.

@property
requires_manual_download = (source) ¶

overrides fiftyone.zoo.datasets.ZooDataset.requires_manual_download

Whether this dataset requires some files to be manually downloaded by the user before the dataset can be loaded.

@property
supported_splits = (source) ¶

overrides fiftyone.zoo.datasets.ZooDataset.supported_splits

A tuple of supported splits for the dataset, or None if the dataset does not have splits.

@property
tags = (source) ¶

overrides fiftyone.zoo.datasets.ZooDataset.tags

A tuple of tags for the dataset.

def _download_and_prepare(self, dataset_dir, scratch_dir, split): (source) ¶

overrides fiftyone.zoo.datasets.ZooDataset._download_and_prepare

Internal implementation of downloading the dataset and preparing it for use in the given directory.

Parameters
dataset_dir	the directory in which to construct the dataset. If a `split` is provided, this is the directory for the split
scratch_dir	a scratch directory to use to download and prepare any required intermediate files
split	the split to download, or None if the dataset does not have splits
Returns
tuple of	dataset_type: the `fiftyone.types.Dataset` type of the dataset num_samples: the number of samples in the split. For datasets that support partial downloads, this can be `None`, which indicates that all content was already downloaded classes: an optional list of class label strings