Create Dataset

On the Main Dataset page click on + Create New Dataset.

Enter the dataset Name and Description and click Create

Dataset Structure

A dataset is a collection of images.
Images may be distributed in groups. Each Dataset can have 0 to many groups
Each dataset can have zero to many annotations sets.
An annotation set may have many annotations,
For any annotation set there may be 0 to many annotations per images
The reason for many annotations sets per image is that one user may annotate images in one annotation set while another may audit them and the audited results are in another annotation set.
This keeps every action on a dataset traceable and segregated

List Datasets

On the main Dataset panels, the list of dataset appears a cards:

Create Annotation Set

To create a new Annotation set, click the add annotation set to dataset icon (+ icon in front of the Annotation Sets on the Dataset Card)
Enter the dataset Name and Description and click Create

Delete Annotation Set

To delete annotation set hover over the annotation on the dataset card.
This will show edit and delete icons
click the delete to delete annotation set
All the annotations in the annotation set will be deleted
The images associated with the annotations are not deleted (as these are associated with other annotations also) . to delete images delete the dataset or delete individual images in dataset

Copy Dataset

Copy dataset allows a virtual copy of the datasets (images are not physically copied)

Copy can be done using filters. This will be explained in the Gallery article.

Under the utility dropdown, select Copy Dataset
The Copy Datasets dialog will appear as below

Source Dataset is automatically populated with the selected dataset. the user can select another project, dataset, or annotation set.
Destination Dataset must be from the current project, which is can be different from the source project. They must have an annotation set selected. Users can have multiple destinations by clicking on the add button next to 'Destination Dataset' The percentage input box indicates the number of images that destination will receive. The total percentage from all destinations must be 100%.
Confidence filters the annotations from the source annotation set between the lower threshold and upper threshold.
Start Date and End Date filter the images be creation date.
Select Labels filter images by the selected labels in this dropdown.

Copy Annotation Sets

To copy annotation from one dataset to another please refer to Copy Multiple Annotations to a Dataset

Upload Dataset

To open the upload dataset dialog, select Upload Dataset under the utility dropdown
The Upload Dataset Modal will appear as below. It allows users to upload single images, multiple images, or entire datasets.
Select an image or group of images by drag and drop into the dialog, or manually select from local directory.
Users can also select a custom import type for their desired dataset.

Images Folder

Upload a group of images by selecting an image folder. All images in the folder will be uploaded to this dataset.
This does not import annotations, so destination annotation set is not necessary.

Darknet

Upload using the darknet format, which requires uploading directory containing 2 sub-directories, 'images' and 'label'.
The 'images' sub-directory can contain more sub-directories, each with their own images
For each sub-directory under 'images', 'label' must contain sub-directories with the same name.
For each image, there must be a .txt label file with the same name as the image.
Label file can contain multiple lines of the format.
```
cls x y width height
```
Cls is the ineteger representing the class name.
X and Y represent the top-left coordinate of the bounding box.
Width and Height of the bounding box.
This option imports annotations, so destination annotation set must be provided.

Darknet Segmentation

Similar to Darknet import, but label files contain coordinates of the segmented mask. The format is as below
```
cls x1 y1 x2 y2...xn yn
```

LabelBox

Select a LabelBox JSON file as source file.
This option imports annotations, so destination annotation set must be provided.

Multiple Videos

For CVAT import, the folder must include videos and label files (JSON) with the same name as the video.

Label file format contains JSON with a key-value pair for every frame. The value for one bounding box is in the format below:

"frame_num": [{"category_id": 1, "category_name": "cat", "bbox": [100, 200, 100, 200], 
"bbox_mode": "BoxMode.XYWH_ABS", "is_seg": 0}]

For segmentation bounding boxes, the format becomes

"frame_num": [{"category_id": 1, "category_name": "cat", "bbox": [100, 200, 100, 200], 
"bbox_mode": "BoxMode.XYWH_ABS", "is_seg": 1, "segmentation": [[100, 200, 100, 300, ...]],
"segmentation_mode": "Polygon" }]

For non-CVAT import, only the videos are required
Destination annotation set is required for CVAT import.

Single Video

Similar to Multiple Videos, but only requires a single video.
User can input an FPS which controls how many frames per second of the video to extract.
This does not import annotations, so destination annotation set is not necessary.

After clicking 'Start Folder Import', the dataset will display a status bar indicating the import progress.

Export Dataset

To open the Export Dataset dialog, select Export Dataset under the utility dropdown
The Export Dataset dialog will open. It allows users to export then download an annotation set to their preferred format.
Selecting type Detection will allow users to choose between 3 formats: Darknet, LabelBox, and eIQ-Portal.
Selecting type Segmentation will allow users to choose between 3 formats: ModelPack Segment, Darknet Segment, and Labelbox Segment.
Select an annotation set as the source to export from.
After clicking 'Export', the dataset will display a status bar indicating the export progress
Download the exported dataset using the download button at the bottom right of the status after complete