Voxel51 <> Segments.ai integration

Machine learning models are only as good as the data they’re trained on. The common saying “garbage in, garbage out” perfectly captures this reality. Even the most sophisticated algorithms will fail if your data is noisy, biased, or incomplete.

Ensuring high-quality datasets requires rigorous data curation. However, as datasets grow larger, it becomes increasingly difficult to identify edge cases and errors. This is where tools like FiftyOne by Voxel51 come into play.

AI/ML teams use FiftyOne to visualize, curate, manage, and QA data, and automate the workflows that support enterprise machine learning. FiftyOne supports various data types like images and point clouds and allows you to filter, slice, and interact with your data both programmatically with Python and in the FiftyOne App interface. It’s particularly effective in helping you pinpoint problem areas in your data and model performance.

However, when you discover new edge cases in your data using FiftyOne, you often need to annotate additional samples for model retraining. That’s where Segments.ai now integrates seamlessly with FiftyOne to close the loop.

FiftyOne helps you pinpoint samples that need annotation, then integrates with the labeling tools you love. That’s where Segments.ai steps in with an official plugin that lets you seamlessly request annotations directly within the FiftyOne interface. This integration eliminates the need to switch between tools and simplifies workflows when refining datasets.

Combining FiftyOne’s data exploration capabilities with Segments.ai’s annotation workflow creates a powerful toolchain for data curation. By connecting exploration, annotation, and quality control in one smooth process, you significantly reduce friction and errors in your dataset, leading to better datasets and faster iteration cycles.

Whether you want to add samples to an existing dataset or create a new one, you can do this without leaving FiftyOne. This streamlined process saves time and reduces manual steps, helping you quickly address data issues discovered during exploration.

You can easily select samples from your dataset in FiftyOne and send them directly to Segments.ai for annotation by running the “request annotations” operator.

Once your team completes the annotations on Segments.ai, you can immediately pull those annotations back into FiftyOne for review. Simply make a new release and run the “fetch annotations” operator.

Quality control is essential. If you spot errors in the annotations, you can create issues directly in Segments.ai or from within FiftyOne. This lets you provide detailed feedback on problematic samples, which annotators can then address one by one.