Method forward

Required Inputs: - objs: LongTensor of shape (O,) giving categories for all objects - triples: LongTensor of shape (T, 3) where triples[t] = [s, p, o] means that there is a triple (objs[s], p, objs[o]) Optional Inputs: - obj_to_img: LongTensor of shape (O,) where obj_to_i

(
    self, 
    objs, 
    triples, 
    boundary,
    obj_to_img=None,
    attributes=None,
    boxes_gt=None, 
    generate=False,
    refine=False,
    relative=False,
    inside_box=None
    )