How are the original `gt_density_map_adaptive_384_VarV2` folder and the corresponding `annotation_FSC147_384.json` file in the FSC147 dataset generated?

[cripsgreen]

Hello, I would like to ask a question. When using your own dataset, do you need to convert it from the COCO format to the FSC147 dataset format? In addition, how are the density-map folders such as gt_density_map_adaptive_384_VarV2 generated? (I noticed that the original FSC147 dataset already contains similar files.) Also, how is the corresponding annotation_FSC147_384.json file produced? Thank you very much!

[cripsgreen]

Sorry to bother you again—one more question. If I create my own labeled dataset, does it have to follow the FSC-147 format where each image contains only one category label? Is it not possible to use multiple object categories per image, like in the COCO dataset?