How to add custom augmentation with lazy config?

[eljirg]

Hi,

I am trying to do transfer learning using one of the new baseline model which is using the new py config.
I have been successful in modifying some of the lazy configs, however, I am stuck when trying to add custom augmentation.

Previously (when using the YAML config), I am using the build_detection_train_loader accompanied with a mapper function.

In this lazy configs sample (https://github.com/facebookresearch/detectron2/blob/main/tools/lazyconfig_train_net.py)
I see that:
train_loader = instantiate(cfg.dataloader.train)

in which the mapper would be in cfg.dataloader.train._target_

I tried to do:
cfg.dataloader.train._target_ = build_detection_train_loader(cfg, mapper=custom_mapper)

However, since the cfg is using the new format, it can't work as the function itself rely on the old config format.

Any advice on how to add custom augmentation with this new format?