Current methods that generate a scene graph of a given image create an overdense graph regarding the number of objects and relationships found in the image, making the graph less effective in describing what is relevant in the image and, consequently, in downs... ...
