could you share the scripts spliting coco datasets into base and novel class datasets? and the contents of 'concepts.txt' file? Thanks advance!

[Sam]

[Yiwu Zhong]

The dataset was provided by OVR paper (Open-vocabulary object detection using captions). You could look into their script for splitting COCO dataset. concepts.txt is simply a file with each row as a class name. See the instruction about creating concept embedding for more details.