These results were obtained entirely automatically, using discriminative mode seeking, on the MIT Indoor 67 dataset. For each Indoor 67 category, we show (on this page) average images of the top 5 detections for each of the top 20 elements. Follow the link for each category to see the full ranking of all 200 elements used in the final classifier. Each row on these linked pages shows the top 20 detections from the target class for a single element. Rankings for the elements are determined using the total weight that the final SVM classifier gave to the element. Again, the linked pages are pretty big--approximately 4,000 patches on a single page.