How is the transformation of annotations in the original dataset into hierarchical JSON files done?

[tslyellow]

For example, in the FSOD dataset, the annotations downloaded from the FSOD official website only include fsod_test.json and fsod_train.json. However, in the annotations downloaded from the link you provided, there are six JSON files:
fsod_test_fixed_l1.json,
fsod_test_fixed_l2.json,
fsod_test_fixed_l3.json,
fsod_test_miu_fixed_2more_miu_final.json,
original_fsod_test.json,
original_fsod_train.json.
Which script was used to convert fsod_test_fixed_l1.json, fsod_test_fixed_l2.json, and fsod_test_fixed_l3.json?
Could you please inform me?

[OatmealLiu]

Hi. We annotated the L1 level of FSOD dataset to get a 3-level semantic hierarchy (GT). The L2-/L3-level class names are from the orginal FSOD dataset. We simply tidy them up into a unified JSON file with our newly annotated L1 level. Since this process is not related to the method proposed in this work per se, we do not include the simple processing code.