From left to right in the first row, we show the input image, the initial segmentation (random colors), and the final generated depth map. Clicking on an image will display it in full size.
The second row shows the back-and-forth video sequence next to the anaglyph image.
This is an example of a limitation of our method: since the tram spans a significant depth range, modeling it using a generalized cylinder introduces strong errors in the depth, which are particularly evident in the parallax sequence. The errors are less noticeable in the still anaglyph image.