Run Inference with Standard Models

Description

With this workflow, Nucleus allows you to run inference with standard computer vision models in order to generate a set of baseline predictions. Generating such predictions can be useful when wanting to benchmark your model performance against a standard benchmark. Alternatively, you can use these model predictions to help intelligently sample data before sending it to be annotated.

The standard models currently supported are:

  • MSCOCO-trained object detection using the EfficientDet D7 architecture
  • MSCOCO-trained panoptic segmentation using the detector2 architecture

Pre-reqs: Dataset

Steps

  1. Go to the Models page by clicking on "Models" in the left sidebar menu.
  2. Scroll to the section titled "Public Models" and click on the desired model (in the video below, we choose object detection).
  3. Once on the model detail page, click the "Run Model" button in the top right hand corner.
  4. In the menu, select the dataset name, and optionally slice name over which you want to run model inference.
  5. After submitting the inference job, you can track the progress in the Job Progress Hub
  6. When the inference job is finished, you'll be able to visualize the predictions in the nucleus grid!