Maps and Geospatial Queries, Simplified Data Upload, Upload Annotations via UI, Free Sale Nucleus access for Education, New Guide: Stop Losing Sleep Over Edge Cases, Introducing Scale Validate, Send Videos to Annotation
We are introducing maps and geospatial queries to Nucleus as a long-requested feature. When uploading any data set with location data attached, Nucleus now automatically plots the data on a map, visible in the “Charts” page. Users can explore the geographic distribution of the data and easily query a subset of their data by drawing a rectangle on the map.
[Explore a dataset with geospatial data →] (https://dashboard.scale.com/nucleus/ds_bwkezj6g5c4g05gqp1eg?explorer_display=image&display=insights&utm_medium=email&utm_source=hubspot&utm_campaign=202209-nucleus-product-update&utm_content=email-1&utm_funnel=awareness)
We have rebuilt the way to upload data to Nucleus using the UI. Users can select between local upload, ingesting from a labeling project or using a sample dataset to get started.
Previously, uploading annotations to Nucleus was possible only using the API interface. With our new data upload wizard, users can now use the web UI to upload labels in COCO and PASCAL format. This is super helpful to get started on Nucleus quickly.
We are excited to share that Scale Nucleus is now available for free for education! Many breakthroughs and advancements in AI happen within university research labs. At Scale, our mission is to serve the world’s most AI-forward companies and practitioners. If you are a student, professor or researcher working in computer vision, you are eligible for a free education account.
We have recently published a new guide on how to deal with long tail data distributions in machine learning, using Nucleus.
We have recently released Scale Validate in Early Access. Scale Validate introduces structured testing for ML models. It allows users to set up tests on mission-critical scenarios to track progress over time and catch regressions. Core to any robust MLOps strategy, Validate helps users to pick the best model through efficient model comparisons and scenario-level insights.
We now support video tasks when sending data directly from Nucleus to labeling. If you are working with video data in Nucleus, simply create a slice of scenes and send these to get labeled with a single click, linking the data to a previously created annotation project.
We have updated the color scheme for the confusion matrix on the Charts page in Nucleus, which improves discoverability of cases where your annotations and model predictions disagree.
We are adding collaboration features to Nucleus, which will enable users to comment on individual data items, tagging colleagues and resolving issues. Comments will help ML teams to quickly see what’s new, flag problems and solve them all without switching platforms or keeping track of things in external spreadsheets or tools.