A dataset is a collection of images, point clouds, annotations, and predictions relevant to your use case. It is continuously updated with new annotations whenever tasks are completed within a project linked to the dataset. You can learn how to create a new dataset in the guide Creating a new dataset .

A chunk is a subset of data (i.e., a group of items) and is useful when grouping, e.g., annotation errors or prediction mistakes of similar characteristics. Apart from the grouping itself, chunks enable users to send items to the annotation workforce for correction, along with instructions for what to correct.

In Dataset Exploration, items are represented as individual thumbnails in the gallery.

For object detection datasets, items are object-centric, meaning that each item represents an individual object in one frame, such as a car, pedestrian, or a traffic sign. For sequence data, each item consists of all frames where a specific object is labeled.

For segmentation datasets, items represent a frame.