How about scene graph generation in other tasks like video captioning or storytelling

[ltp1995]

Hi, i recently attempt to make a graph generation of VIST dataset in storytelling task, which has another vocabulary(9837 words totally) different from coco dataset. Then a sg.json file is produced by the create_coco_sg.py you provided, however, the generated (attribute, object, related and subject) dicts in sg.json file are really bad that are fail to capture these relations, so I wonder if the files like create_coco_sg.py and process_spice_sg.py can extend to deal with other datasets, not only coco dataset, and how can I generate scene graph well in other datasets of other tasks.