Today, we’re proud to officially introduce two key product improvements to the Figure Eight platform to expedite the creation of high-quality training data for our customers. We’ll start by walking you through our new image annotation solution and follow up with a brand new feature we’re calling Figure Eight Workflows.
Figure Eight Image Annotation Tool Enhancements
We’ve seen an explosion in image annotation jobs in the past few years. More and more organizations are leveraging machine learning to identify everything from products on shelves to advertisements in video to pedestrians on the road. Our suite of computer vision solution has powered all these and more, but we’re always looking to make our core platform offerings better. Here are our newest enhancements, all aimed at making creating the best computer vision training data a little bit easier:
- We’ve added the ability to create test questions on bounding boxes, polygons, dots, and lines, along with the labels applied to those shapes. This allows customers to test contributors before entering a job as well as while they work to ensure their judgments remain accurate across all parts of the job.
- All shapes and their labels can now be aggregated, so customers can collect multiple contributor annotations per image and aggregate them into the best overall result
- Self-serve ontologies allow customers to create their own ontologies for image annotation, so they are easily customizable and can be changed from job to job
- A new image annotation graphical user interface (GUI) for job design, along with pre-built templates, to make it easy to quickly set up new jobs
If you’d like to try these tools out, please take a moment to read our Success Center documentation to get the most value from these upgrades:
- Guide to Running a Bounding Box Job with Labels
- Guide to Polygon Job Design, Test Questions, and Aggregation
- Guide to Running a Dots Job with Labels
- Guide to Polylines Job Design, Test Questions, and Aggregation
Figure Eight Workflows
Creating high-quality training data sometimes requires multiple steps and multiple annotation tasks, especially for the complex machine learning problems our customers look to solve. That’s why we created Figure Eight Workflows.
With Figure Eight Workflows, our customers can link multiple Figure Eight jobs to each other, automatically route data, and run a sequence of annotation flows together. In effect, this helps break down complex annotation tasks into simpler jobs with discrete steps to achieve higher quality output and lower their potential external crowd cost.
We’ve gotten great feedback from our customers about this feature already. Here’s why they like it:
You cam automate workflows of any complexity with an easy to use interface
- Simple UI driven tool means making a workflow is painless
- Configure data routing rules between jobs for fully customized workflows
- Automatically stream data and kick-off jobs when data flows in
- Aggregated results are generated at the end for the entire workflow
It improves quality and efficiency
- Running multiple simpler, targeted jobs means less cognitive load for labelers
- Smaller, simpler jobs have been shown to increase both efficiency and quality. With less options and steps, contributors can become expert in discrete tasks and excel.
- Since jobs run in parallel, it increases throughput. There’s no need to wait for a job to finish to annotate specific data from that job later in a workflow.
Want to learn more about these improvements or give them a try with your next Figure Eight project? Reach out to your Figure Eight Customer Success Manager or our Figure Eight Support Team for more information. Thanks!