How to get a metric during training, for example 'CocoMetric' from the validation and training dataset

[EvgeniyBerdyugin]

Thank you very much to the developers for your work. I would like the quality metrics at the validation stage to be calculated on both the validation and training datasets. I tried passing a similar list to val_evaluator:

val_evaluator = [
    dict(
    type='CocoMetric',
    ann_file='data/masks_coco/' + 'val.json',
    metric='bbox',
    backend_args=None),
                 dict(
    type='CocoMetric',
    ann_file='data/masks_coco/' + 'train.json',
    metric='bbox',
    prefix='train/',
    # data_prefix=dict(img='train/'),
    backend_args=None)
                 ]

However, the metrics on the training dataset are zero, as if the annotations were loaded as training ones and the images as validation ones. Apparently it is necessary to convert val_dataloader, but unfortunately I did not find a way to register two datasets in the documentation. Or maybe I need to write custom hooks here?