Natural Language Search for 2D and 3D Training Data with CLIP in Nucleus

Nucleus is already an incredibly accessible way to curate your training data, and today, instead of writing queries with precise syntax, you can just type what you’re looking for, such as “a photo of a police car.” If a 2D image is matched with a corresponding point cloud, you can explore that point cloud and its tags to see if it suits the “edge case” you’re investigating, in order to train your model to handle it.

No need to learn syntax: filter your dataset with plain old words Slice and dice your 3D data with only a text search Speed up your dataset creation pipeline, even with challenging data types like 3D Search for scenarios even in unlabeled data with natural language