This task type uses CLIP to do component labeling, and only keep the main object (with the largest volume).