Conditional Random Fields for Image Labeling

Figure 9

Results of [1] on PascalVOC-10 dataset. From (a–e): input image, ground truth, output from [40] (AHCRF + cooccurrence), output from [34] (dense CRF), and output from dense CRF with Potts and cooccurrence terms [1].

(a)

(b)

(c)

(d)

(e)