Detectron2: How to load only weights?

[ashkanaev]

I trying to start new train stage without loading optimizer and scheduler from previous checkpoint. The resume flag is set to false, but these states(optimizer, scheduler) are still loaded.... and i can't valuably configure new stage of training procedure.
How do i configure the train config to load only weights?

[ppwwyyxx]

You can read/write the weight file using torch.load/save and modify the object directly.

[wangg12]

https://github.com/tianzhi0549/FCOS/blob/master/tools/remove_solver_states.py