The linked files and paper descriptions contains different numbers of Object and Attribute Synset

[fake-warrior8]

Hi, I downloaded the linked files and found 2 attributes files, attribute_syssets.json, attribute_synset_values.json, each file has 18720, 5959 items, while the paper shows that there are 16,693 items. Similarly, there are 3 object files, object_synsets.json, object_synset_values.json, object_alias.txt, each has 40154, 5774, 3435 items, while the paper shows 7414 items. why is there a difference?

[MikeWangWZHL]

Thanks for the question! Yes, those are the original files from visual genome; you can find the filtered files here: https://github.com/MikeWangWZHL/VidIL/tree/main/visual_token_ontology/vg

• Objects: https://github.com/MikeWangWZHL/VidIL/blob/main/visual_token_ontology/vg/openimage_classes_all_cleaned_fictional_characters.json
• Events: https://github.com/MikeWangWZHL/VidIL/blob/main/visual_token_ontology/vg/vg_srl_selected_object_synsets_keys_remove_similar0.9.json
• Attributes: https://github.com/MikeWangWZHL/VidIL/blob/main/visual_token_ontology/vg/vg_original_attributes_synsets_keys_cleaned_remove_similar0.9.json

---

The cleaning process are detailed in the Appendix as follows: