The scene graph parsing model pretrained on VG scene graph dataset and finetuned on FACTUAL. Please see 'https://github.com/zhuang-li/FACTUAL' for details.