The image annotation templates available on Datax Task Manager include bounding box image annotation, point image annotation, image categorization, and image sentiment analysis. Read on to find out how you can use these task templates to get the data you need.
Image annotation is the perfect way to get your images labeled objectively according to the items visible in them. For example, if you're putting together a data set to train an AI to recognize dogs in images, you might use our image annotation template and instruct contributors to identify all the dogs in a series of pictures by drawing a bounding box around them.
Image annotation also works for gathering more subjective data—marketing professionals may find this especially useful. Using the image annotation template, you might upload a series of ads ask our contributors which part of the image strikes them as the call to action, or even whether the ad appeals to them.
There are two ways to get started creating an image annotation task in Datax Task Manager: either choose New Task on your dashboard then select the image annotation task template you'd like to use, or enter an existing task and add an image annotation question under the Design tab.
Under Data & Resources, create an image resource group and upload the images you need annotated.
You can keep working on other parts of the task while waiting for the images to upload — just don't exit to the task list, refresh the page, or leave the website.
Under Design & Resource, select the question, add an image annotation question and select the appropriate image resource group for the question.
Before you leave the question design stage, make sure to check "No item" text override and set a text override if needed. For example, if your task involves contributors identifying all the trees in an image, set a simple phrase like "No trees" for images that don't have any trees.
If your task is potentially confusing, you might benefit from including illustrated instructions. Namely, make a collage of 3-5 images displaying the way you'd like your images labeled. One of these examples should be straightforward and the rest should involve more controversial images so that contributors can be sure of what to do if they come across a confusing image.
Use a simple illustration to showcase disputable cases.
You're almost done: the last order of business is to configure contributor requirements and distribution options. Check out our guides on these topics for some help.
Publish your task and wait for the responses to come in!
Reviewing responses is a crucial step in the data collection process. While you're doing this, you can change the dot color or box color by hovering your mouse over an image, and zoom in by clicking on an image.
When you reject a response, be sure to select the reason you did so! Datax analyses these rejection patterns to improve our contributor app and ensure better responses to your tasks in the future.
Tip: use keyboard shortcuts to speed up the response review process.
Our contributors are human! The trick to crowdsourcing data effectively is designing your tasks to engage contributors, maximizing the number of useful responses you'll get.
Still have any questions? Reach out — our team is delighted to help!