Freeze the pre-trained weights and add some additional module(s)

[seulkiyeom]

Hi, I'm a big fan of mmseg.

I'd like to

1. add some module(s) based on the pre-trained model and
2. train the model while freezing the pre-trained weights (except for the added modules) to update the modules only.

I'm currently implementing based on

mmseg==1.0.0
mmcv==2.0.0
mmengine==0.7.3
mmcls==1.0.0rc6
mmdet==3.0.0

Because in general, training and test procedures should be run on Runner (from mmengine) (e.g. runner.train() and runner.test()), I'd kindly ask how to add modules on our pretrained model and train/test the modified model with freezing.

Is there any specific function I can try on it? Best

[daeunni]

@seulkiyeom Hi, I have the same problem. Did u address it? :D

[Meoooww]

@seulkiyeom Hi, I have the same problem. Did u address it? :D