1) Predictions: Model based pre-labeling
Why label data from scratch if your model can help? Labelbox now allows you to import model predictions as pre-labels via the Prediction API.
The result is a significant reduction in the time and cost of human labor needed for labeling. The model generated pre-labels are imported into Labelbox via the Prediction API and then presented as an initial recommendation for each labeling task. This transitions the focus of the labeling teams from zero-to-one-labeling tasks to QA. You can expect up to a 5x increase in labeling speed for bounding boxes and classification tasks by operating a semi-automated labeling workflow in this manner.
(2) Labelbox Workforce
Labelbox now offers a fully managed and dedicated workforce to all customers. With transparent pricing and a dedicated project manager, your team can safely rely on Labelbox for a fully integrated training data solution. This workforce offering is available in addition to the workforce providers already available on Labelbox, including CloudFactory, Daivergent, Cogito, Edgecase and iMerit. As a Labelbox customer, you can connect to one or more of these workforces right inside your projects.
With the Labelbox Workforce, machine learning teams have access to a dedicated labeling team with whom they can collaborate and train for their use cases. Together with pre-labeling and automated quality control features, Labelbox customers are now spending much less on human labor and creating high-quality training data at scale.
(3) Real-time workflow
Labelbox Real-time enables teams to use a concurrent labeling task distribution and queue system in the loop with production applications. In other words, raw data can now be streamed directly into Lablebox to be labeled by the labeling team that is working 24/7. The labeled data is sent back within a specific time (as quickly as a few minutes!) to the application. Meet stringent SLAs and tight turn arounds on your production system with Labelbox Real-time. This feature is available to all users starting today.
(4) Releasing reservations during logout
When you “Start Labeling”, Labelbox automatically reserves 10 data rows for you. These reservations expire after 90 mins of inactivity. For smaller projects, reserving 10 data rows isn’t ideal. Now when you log out, Labelbox releases reservations across all your projects. In June, we plan to reduce the reservation count to 3 for all users.
(5) Support for TIFF files
We now support TIFF files. Read more: https://support.labelbox.com/docs/uploading-datasets-1#section-tiff-files