About coco in Project-MONAI/tutorials/tree/main/detection

[Dana136484]

When calculating the validation metrics after several rounds of training, coco is calculated for all classes, which results in the optimal metric being negative. But I used my own dataset for training, and the category labels of all bounding boxes in a training set only contain some classes, so only the category labels of the bounding boxes in ground truth need to be calculated.How should I modify it?Thanks!
all classes:


The bounding box category labels for my own data: