Image Segmentation

Segment instances on image datasets using the Universal Data Tool

Setup the Dataset

Navigate to udt.dev and click "New File"

Click "New File" on udt.dev

Then select the Image Segmentation button from the Setup > Data Type page.

You can configure the Image Segmentation to create the right interface for your dataset. Use the Setup > Preview button to see your interface against either an example image or a sample from your dataset.

Image Segmentation with Bounding Box Classification

Configuring Regions and Labels

The Image Segmentation interface allows different types of regions, the options for regions are:

• bounding-box

• polygon

• point

Each region can any number of labels, which can be configured under "Available Labels"

Import Data

You can use any of the following methods to import image data. If you're just getting started, you can quickly create a dataset using the COCO Images method!

• Import COCO Dialog
• Import from Google Drive
• Import from AWS S3 Bucket
• Import from List of URLs
• Import from CSV or JSON
• Upload or Open Directory

Label your Data!

Use the Label tab to label your data. Look at the Collaborative Labeling Guide to label with others.

Export and Use

You can download your data using the download icon at the top.

Download your data in CSV or JSON format to use the annotations

You can use the Universal Data Tool Converter to convert UDT files into PNG masks, or other formats that are helpful for machine learning datasets.

It's very common to convert into PNGs