Auton Lab is developing an AutoML pipeline that automates the process of featurizing data and training for many different model classes. The output of the system is a number of candidate pipelines which are then evaluated to find the best pipeline for a given prediction task. Currently, AutonML supports image and video data, but produces a small number of candidate pipelines because there are a limited number of image and video primitives included in D3M. However, there are extensive existing CV featurizations in toolkits like skvision and opencv that could be used to improve the results. We will start by considering about a dozen datasets, evaluate with the existing RESNET primitives, and then implement some new primitives and filters, re-run autonML, and compare the results. For a particular application context, we have been developing AI to detect instances of stenosis from coronary angiogram videos. Currently, models are built using information from individual images, but the goal would be to reuse all the labels for video data and compare the featurizations of video and image information.