tlc.integration.super_gradients.datasets.detection_dataset

Detection Dataset for SuperGradients.

Module Contents

Classes

Class

Description

DetectionDataset

API

class DetectionDataset(
table: Table,
image_column_name: str = 'image',
label_column_name: str = 'bbs.bb_list.label',
**kwargs: Any,
)

Bases: super_gradients.training.datasets.detection_datasets.detection_dataset.DetectionDataset

A dataset for training SuperGradients detection models, populated from a 3LC Table.

Note

SuperGradients models expect the labels to be contiguous from 0 to num_classes - 1. If the label value map keys are not contiguous, a mapping is applied for the DetectionDataset and the reverse is applied in the DetectionMetricsCollectionCallback. In such cases, model predictions outside of 3LC will have contiguous labels, which are not consistent with the keys in the 3LC Table.

Parameters:

  • table – The table to load image paths and ground truth boxes from.

  • image_column_name – The name of the column containing the image paths.

  • label_column_name – The value path to the ground truth labels.