Table 3. Results for the Scribble dataset. We compare single-scale and multiscale inference and report results for a selection of competitive methods that were not incorporated in our framework. On this challenging large-scale dataset, multiscale inference was repeatedly superior to single-scale. QPBO (long) denotes a singlescale, iterative application of QPBO-I; this highlights the advantage of multiscale inference, as even when QPBO-I was run exhaustively it came short compared with multiscale inference.
...read more