Figure 4From: Scene Segmentation with Low-Dimensional Semantic Representations and Conditional Random FieldsIllustrative labeling results on (a) Sowerby and (b) Corel. Columns from left to right: (a) input images; (b) first stage LRC, mapped to pixel level using the nearest-patch-center method; (c) LRC/LRC with MRF smoothing; (d) LRC/CRF with oversegmentation; (e) hand-labeled ground truth.Back to article page