We will show how you can achieve the concept of “Operation Vacation” for the models you create, and make sure that the model is testing the subsets that you actually care about with Scale’s latest product, Nucleus. Using nuScenes 2.0, a multimodal dataset for autonomous driving, we will demonstrate how you can easily debug model performance and automatically refine your model. In the process, we’ll also dive into Nucleus’s features to show how to curate sub-datasets and edge cases easily with custom metrics, image similarity search, and auto-pivot, automatically augmenting the data collection to accelerate machine learning training process.