With Smart Sample, Nucleus makes it easier and faster than ever to build the best datasets for labeling, (re-)training or testing. Nucleus already has powerful search and filtering capabilities, but up until now still required users to pick the data they need manually from search results. With Smart Sample, users can now simply define the filter and sampling criteria for their target data subset and Nucleus automatically generates the best fitting data slice.
Upload a dataset to try Smart Sample →
We have added a new public dataset to Nucleus. COCO-Stuff by Holger Caesar, Jasper Uijlings and Vittorio Ferrari augments all 164K images of the popular COCO dataset with pixel-level annotations. These can be used for scene understanding tasks like semantic segmentation, object detection and image captioning. On Nucleus, we also predictions of a panoptic segmentation model. Nucleus makes it easier to use COCO-Stuff as you can search for specific annotations, predictions or using natural language descriptions (e.g. images of boats in a harbor).
Explore COCO-Stuff on Nucleus →
If you are working on improving ML model performance within a team, it is important to stay on the same page when discussing specific data points, annotation issues or model errors. That is why ML teams of all sizes use Nucleus to collaborate seamlessly on their datasets and share direct links to specific issues. We have made it easier to add your team members to Nucleus to accelerate the process of sharing direct pointers to datasets, models, annotation tasks and more.
Invite a team member to Nucleus →
We are rebuilding the process of uploading data, annotations and model predictions to Nucleus from scratch. Coming up, we will release a massively simplified UI to ingest raw data from a variety of sources including direct connection to cloud buckets. Furthermore, we are newly adding the ability to upload annotations and model predictions using the web interface. Stay tuned!