![](https://assets-global.website-files.com/62e939ff79009c74307c8d3e/63e37872910dd455a8da1cfc_External%20Labelling%20Workforce%20Options.png)
Why is an External Labelling Service Needed?
As our world becomes increasingly data-centric, data quality has solidified itself as an essential metric for success with the computer vision model training pipeline. A large part of maintaining data quality is careful data annotation. However, data annotation is a time-consuming process, so much so that data labelling and curating can take up the largest portion of time in the whole machine learning process. This cost of time can be worsened by increasing annotation complexity, such as the tedious work required to create polygons that outline objects well. Additionally, if annotation errors are made, this can have serious implications on the ability for the computer vision model to be able to learn well, given that the ground truth labels might also be incorrect.
With an external labelling service, one can offload this important but tedious manual work to a larger, more experienced workforce with prepared workflows in place to provide quality annotations that are cross-verified and reliable. This allows the user to focus on the more exciting and conceptual aspects of the computer vision pipeline, such as training, iterative experimentation, and deployment.
When Should You Use an External Labelling Service?
An external labelling service is necessary when you have the image data in hand but the quantity of data that needs to be annotated is too much for your timeline, or if annotating that data on your own will be too slow to match required deadlines. For example, annotating video data is particularly tedious because one has to annotate the object in each frame. A relatively short video can contain thousands of image frames. This is far too tedious and time-consuming for a single individual to do for multiple videos. This is a clear case that could benefit from an external labelling service. In another case, if a pipeline is using active learning to iteratively improve upon models that are already being deployed in production. In this case, it is essential that the whole process is streamlined and rapid so that the newly updated model can be used to improve the performance as compared to the current version. If the process takes too long, performance at the production level may suffer, and other changes may have already occurred in that timeframe beyond what the new version of the model was being trained to deal with.
If your projects sound like they could benefit from the help of external labelling, doing so will undoubtedly improve the quality and efficiency of your computer vision pipeline.
How Does Our External Labelling Service Work?
Our external labelling service starts with users requesting for help from the Annotator page. The request is a simple three step process in which one describes their project, uses our Annotator to annotate 8 images as samples, and outlines the budget and timeline in which the annotation job should be completed.
In terms of pricing, our minimum price is 300 USD, and scales with the number of images needed for annotation, the average number of annotations per image, whether the type of annotation is for bounding boxes or segmentation masks, and the required annotation pace.
During the labelling job itself, Nexus’ external labelling infrastructure ensures that all annotation work is done on the platform, so user project data is secure and important private information is not made available to the labellers. External labellers will have limited feature access restricted to the Annotator and basic project statistics information on the homepage during the annotation job. Importantly, only images that are assigned for external labelling will be accessible to the labellers. All access will be removed after the job has been completed. The platform also supports tagging of assets to facilitate easy quality control and review.
For a more detailed breakdown of how to use our external labelling service on Nexus, please check out our tutorial here. For more technical and in-depth details, check out our documentation! Additionally, if you have specific requirements that don’t meet the granularity specified in our onboarding page, feel free to contact us directly to inquire about how we can best fulfill your annotation needs.
Our Developer’s Roadmap
Our External Labelling Service is another part of Datature’s effort in helping to speed up and make the computer vision pipeline more efficient. We will therefore continue to refine our labelling services with greater granularity of control, such that data can be split up for different groups of annotators, as well as provide safety features to restrict the ability of external annotators to access resources not relevant to their annotating job.
Need Help?
If you have questions, feel free to join our Community Slack to post your questions or contact us about how the External Labelling Service fits in with your usage.
For more detailed information about the External Labelling functionality, customization options, or answers to any common questions you might have, read more about the annotation process on our Developer Portal.
Build models with the best tools.
develop ml models in minutes with datature