Description

You can run Scale's first Automotive Foundation Model (AFM-1) on all of your data within Nucleus! This will allow you to run model inference with flexible taxonomies using panoptic segmentation, semantic segmentation and object detection models. This is perfect for auto-labeling your data with a state of the art zero shot model! The model results will be ingested into Nucleus and fully queryable through our query engine.

Pre-reqs: Dataset upload

Steps

1. Open a Nucleus dataset
2. On the dataset overview page, select the Run AFM-1 card.
3. Within the opened modal, you'll need to define
   1. (optional) A slice: In case you don't want to run inference on the entire dataset, but only a subset of data.
   2. The model type: Pick between panoptic segmentation, semantic segmentation and object detection.
   3. The taxonomy: AFM-1 support free vocabulary taxonomies. Therefore, you can either select suggested classes or enter desired ones using free text.
   4. Run name: Enter a unique model run name which you'll later use to view and query the inference results.
4. Click Next and afterwards Confirm to kick off the inference job.
5. Check the job status on the Jobs page.
6. Once complete, check out the Explore view of your dataset in order to query for your results or transition to the detail view to analyze the results in more detail