load pretrained weights

[cyl112233]

Hello,
I would like to replace the model with exp53_SFNetDecoder during training. However, when I load the pretrained weights from:checkpoints/camvid_exp53_decoder29_200_epochs_288_1152_resize_run1
I encounter the following error:File "/Y_L/fifi/RegSeg-main/model.py", line 66, in init
if dic[k].size()==current_model[k].size():
KeyError: 'decoder.head16.stages.0.1.conv.weight'
I understand this means the pretrained weights are not compatible with the new decoder architecture.

But if I don’t use the pretrained weights, the performance is significantly worse.

May I ask: how did you load pretrained weights when changing the decoder to SFNetDecoder?
Is there a recommended way to partially load weights (e.g., only the encoder)?

Thank you very much.