Both files are used to pre-process the MSCOCO validation dataset for our pipeline. Files need to be placed under the dataset folder in the following structure, while args_cmd.dataset_dir is datasets, dataset_raw_data is coco_val_sets, dataset_obj_db is obj_db_clean.json or corr_obj_db_clean (only for Paired Object Insertion) and dataset_scene_db is scene_db_clean.json:

├── datasets
│      └── coco_val_sets
|      └── obj_db_clean.json
|      └── scene_db_clean.json
|      └── corr_obj_db_clean.json