Pull the Plug? Predicting If Computers or Humans Should Segment Images Supplementary Material

This document supplements Section 3 of the main paper. In particular, it includes the following: 1. Our segmentation initialization system illustrated by example (supplements Section 3.2). 2. Results illustrating the versatility of our segmentation initialization system for three segmentation tools (supplements Section 3.2). 3. Results illustrating the performance of our predicted initializations against two initialization baselines (supplements Section 3.2). 4. A variant of our initialization system for interactive segmentation tools which embed a shrinking bias (supplements Section 3.2). 5. Parallel results to Figure 3 in the main paper showing the benefit of our segmentation initialization system when using simulated human input (supplements Section 3.2). 6. Our fine-grained segmentation system illustrated by example (supplements Section 3.3). 7. Parallel results to Figure 5 in the main paper showing the performance of our system when evaluating human effort with respect to number of user clicks (supplements Section 3.3). 8. Methods to collect crowdsourced segmentations (supplements Sections 3.2, 3.3).