Assume that we want to get the annotation of a certain image whose imgname = 'indoor_02018.jpg', we can use cv2 to read instance annotation:

ins = cv2.imread(imagename.replace('.jpg','.png'),cv2.IMREAD_GRAYSCALE)

the corresponding semantic segmentation annotation can be read likewise.

To get the relation annotation whose annotation file name is 'relation.json', run:

rel = Relation()
rel.loadAnnot('relation.json')

To get the relation annotation of a certain image, for example, 'indoor_00002.jpg', run:

relation = rel.loadRelation('indoor_02018.jpg')

To visualize the semantic segmentation map, we can run:

visualize_segmentation(segmentationMap, img)

To visualize the instance map of a certain category, we can run:

visualize_instance(instanceMap, segmentationMap, oriImg)

To visualize relation, run:

visualize_relation(relation, instanceMap, img)

img is the original image.