High-quality training data is crucial for the success of any machine learning model. Having a scalable and systematic way to measure labeling quality is integral to the creation and maintenance of high-quality training data. Labeling quality refers to the accuracy, consistency, and reliability of the annotations produced by human labelers or automated labeling models.
Labelbox’s Performance Dashboard helps teams observe and manage their Labelbox labeling projects. Broken down into three components: throughput, efficiency, and quality, the dashboard provides a holistic understanding of your project’s entire labeling operations. You can learn more about the importance of throughput, efficiency, and quality in our recent blog post.
As of late September 2023, we’re introducing key updates to the Performance Dashboard to improve accuracy, provide additional metrics, and offer more insight and granularity into labeling performance:
Accurately measuring the time taken to label assets is vital to maintaining and understanding labeling quality and efficiency. To address product feedback around how time was logged and accrued after a label had been created, we made improvements to how time is measured and logged during the labeling process.
The above changes help prevent the accumulation of idle time when a labeler is not actively engaged in the labeling process.
The new dashboard will provide more accurate labeling activity metrics based on the changes stated above to calculate labeling time.
With the migration to using batches, the Data Rows tab, and workflows, for more granular control over your labeling operations, we now look at labeling, review, and rework as separate activities:
Labeling time: Displays the total time spent labeling data rows
Review time: Displays the total time spent reviewing labeled data rows
Rework time: Displays the total time spent reworking labeled data rows
Introducing improvements to help your team dig into more specifics and provide more insight into labeling performance:
These new observability improvements help to ensure that your labels are high-quality and were designed to accelerate the labeling process and help cut overall annotation and review costs.
As we roll out these improvements across our user base, we encourage you to leverage the new dashboard for better insights and learn more in our documentation.