data_labeling

Introduction to DeepCell Label

Table of Contents

• Overview
• Load files
  • Drag & Drop files onto label.deepcell.org
  • Supported filetypes
• Use DeepCell Label
  • Guided walkthrough with a demo file
  • Use RGB mode

Overview

DeepCell Label is an browser-based data labeling tool designed for biological images. DeepCell Label allows you to view biological image and create or modify labels on top of them, building or modifying a single-cell segmentation of the image. In this guide, we'll walk through starting up DeepCell Label and the labeling tools available on this platform.

Load files

Drag & drop images on label.deepcell.org

The easiest way to use DeepCell Label is to visit label.deepcell.org and drag & drop your images into the upload file box. DeepCell Label will create a new project for you and redirect to the project URL, formatted like label.deepcell.org/project/{projectID}, where projectID is 12 random characters.

We also offer a selection of test files in a drop down menu on the DeepCell Label homepage, allowing you to explore DeepCell Label without providing your own data. Opening a test file from the dropdown menu will create and redirect you to a new project URL.

We will be working with this demo file throughout the upcoming guided walkthrough. Try downloading it now and opening it in DeepCell Label by dragging and dropping.

Supported filetypes

A project in DeepCell Label consists of a raw image stack and a labeled image stack. DeepCell Label can both load files with only a raw image stack, or a raw image stack and a labeled image stack. If you load a file with only a raw image stack, DeepCell Label will create an empty labeled stack that you can label from scratch.

At the moment, DeepCell Label supports the following filetypes:

• .npz - zipped numpy arrays
  • contains paired raw and labeled image stacks
  • we can package multiple image stacks into an .npz with numpy's savez function, like with the following Python code snippet:

  array_shape = (30, 512, 512, 1)  # array shape is (frames, height, width, channels)
  raw_image_stack = np.zeros(array_shape, dtype=np.uint16)
  labeled_image_stack = np.zeros(array_shape, dtype=np.uint8)
  # important to use X and y kwargs to name the zipped arrays
  np.savez('/path/to/your/file.npz', X=raw_image_stack, y=labeled_image_stack)
  

• .png - loaded as a single raw image
• .trk - a custom file format for DeepCell Label tracking projects
  • consists of a .tar file with a raw image stack, a labeled image stack, and lineage metadata
  • leave an issue on the DeepCell Label repository for more details on working with this filetype

Use DeepCell Label

Want to learn more about the tools available in DeepCell Label? Start with our guided walkthrough of a provided demo file, or load up a test file from the menu on the DeepCell Label homepage.

Curious about visualizing your image channels with a multicolor overlay while you work? Check out our overview of how to switch to an RGB viewing mode, and what changes to expect.