A Symmetric Fusion Learning Model for Detecting Visual Relations and Scene Parsing

Figure 4

Qualitative examples of detecting visual relationships by the baseline method and proposed model: (a) a test image from the VRD data set, (b) the extracting visual relationship triplets by the baseline method, and (c) the extracting visual relationship triplets by the baseline model with the and module. The entity pairs have the same color as the corresponding bounding boxes. indicates the detected error visual relationships. subject-predicate-object represents the detected proper visual relationships different from ground-truth triplets.

(a)

(b)

(c)

(d)

(e)

(f)