how to fix feature encoder weights during SS process

[mengruwg]

during pythorch version new_weight = self.weight.mul(new_mtl_weight)(line 95 in conv2d_mtl.py) self.weight = Parameter(torch.Tensor(out_channels, in_channels // groups, *kernel_size))(line 42 in conv2d_mtl.py) How to load pretrained and fix feature encoder weights during SS process?

[yaoyao-liu]

How to freeze the convolution weights?

If you set mtl=True in the following line: https://github.com/yaoyao-liu/meta-transfer-learning/blob/fe189c96797446b54a0ae1c908f8d92a6d3cb831/pytorch/models/resnet_mtl.py#L165

It means that you're using _ConvNdMtl function, so https://github.com/yaoyao-liu/meta-transfer-learning/blob/fe189c96797446b54a0ae1c908f8d92a6d3cb831/pytorch/models/conv2d_mtl.py#L45 i.e., the convolution weights are frozen.

How to load the pre-trained model?

You may directly load normal checkpoints to MTL models like this: https://github.com/yaoyao-liu/meta-transfer-learning/blob/fe189c96797446b54a0ae1c908f8d92a6d3cb831/pytorch/trainer/meta.py#L69

If you have any further questions, feel free to ask.

[mengruwg]

Thanks for your detailed reply. I can completely understand the experiment now.

在 2021-05-12 17:26:14，"Yaoyao Liu" @.***> 写道：

Reopened #52.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub, or unsubscribe.