by No License
by Unknown License
Ground truth segmented image sequence data is very hard to come by. The difficulty of hand labelling single images is prohibitive enough without venturing to image sequences. One way to obtain such sequences in a cheaper fashion is to make a compromise on the original images. If the images are generated, say, through ray tracing, then the same generative process can be used to create the segmentations. This has the benefit of providing useable data while having the disadvantage that the images are not real-life. However the images do have significant lighting, specular effects etc, so they are more realistic than some other images might be.