core.data.dataset.coco_dataset

coco_dataset

Classes:

Name	Description
CocoDataset	A specialized dataset for managing COCO annotations, enabling downloading, batching,

CocoDataset(name, dataset_version, assets=None, labelmap=None)

Bases: BaseDataset

A specialized dataset for managing COCO annotations, enabling downloading, batching, and merging of annotation files.

This class provides methods for downloading annotations in batches, merging them into a single COCO file, and loading the data for further processing.

Parameters:

Name	Type	Description	Default
name	str	The name of the dataset.	required
dataset_version	DatasetVersion	The version of the dataset to work with.	required
assets	Optional[MultiAsset]	Preloaded assets, if available.	None
labelmap	Optional[Dict[str, Label]]	Mapping of labels for the dataset.	None

Methods:

Name	Description
download_annotations	Download COCO annotations in batches, optionally merging them into a single file.
load_coco_file_data	Load COCO annotation data from the merged annotation file.
download_assets	Downloads all assets (e.g., images) associated with the dataset to the specified directory.
get_assets_batch	Retrieves a batch of assets from the dataset based on the specified limit and offset.

Attributes:

Name	Type	Description
coco_file_path	str | None	The path to the merged COCO annotation file.
coco_data	dict[str, Any] | None	The loaded COCO annotation data.
name		The name of the dataset.
dataset_version		The version of the dataset from Picsellia.
assets		A preloaded collection of assets. If not provided, assets will be dynamically listed.
labelmap
images_dir	str | None	The local directory where image assets are downloaded.
annotations_dir	str | None	The local directory where annotation files are stored.

coco_file_path = None instance-attribute

The path to the merged COCO annotation file.

coco_data = None instance-attribute

The loaded COCO annotation data.

name = name instance-attribute

The name of the dataset.

dataset_version = dataset_version instance-attribute

The version of the dataset from Picsellia.

assets = assets instance-attribute

A preloaded collection of assets. If not provided, assets will be dynamically listed.

labelmap = get_labelmap(dataset_version=dataset_version) instance-attribute

images_dir = None instance-attribute

The local directory where image assets are downloaded.

annotations_dir = None instance-attribute

The local directory where annotation files are stored.

download_annotations(destination_dir, use_id=True)

Download COCO annotations in batches, optionally merging them into a single file.

Parameters:

Name	Type	Description	Default
destination_dir	str	Directory to save the COCO annotation files.	required
use_id	Optional[bool]	Whether to use asset IDs in file paths (default: True).	True

load_coco_file_data()

Load COCO annotation data from the merged annotation file.

Returns:

Type	Description
dict[str, Any]	Dict[str, Any]: The COCO data loaded as a dictionary.

download_assets(destination_dir, use_id=True, skip_asset_listing=False)

Downloads all assets (e.g., images) associated with the dataset to the specified directory.

This method retrieves and downloads all the assets linked to the dataset version. If assets are preloaded, they are directly downloaded; otherwise, the method dynamically lists and downloads them from the dataset version.

Parameters:

Name	Type	Description	Default
destination_dir	str	The directory where assets will be saved locally.	required
use_id	Optional[bool]	If True, uses asset IDs to generate file paths. Defaults to True.	True
skip_asset_listing	Optional[bool]	If True, skips listing assets after downloading. Defaults to False.	False

Side Effects

• Creates the destination_path directory if it doesn't already exist.
• Sets self.images_dir to the destination_path.

get_assets_batch(limit, offset)

Retrieves a batch of assets from the dataset based on the specified limit and offset.

This method is useful for processing large datasets in smaller chunks.

Parameters:

Name	Type	Description	Default
limit	int	The maximum number of assets to retrieve in the batch.	required
offset	int	The starting index for asset retrieval.	required

Returns:

Name	Type	Description
MultiAsset	MultiAsset	A collection of assets retrieved from the dataset.