Closed wanboyang closed 2 years ago
and I retrain the Difnet_LRP by using the default setting.
The performance is lower than the one reported in this paper.
when I use pretrained model DIFNet.pth, some errors reports:
RuntimeError: Error(s) in loading state_dict for Difnet_LRP: Missing key(s) in state_dict: "decoder.layers.0.pwff.layer_norm1.weight", "decoder.layers.0.pwff.layer_norm1.bias", "decoder.layers.1.pwff.layer_norm1.weight", "decoder.layers.1.pwff.layer_norm1.bias", "decoder.layers.2.pwff.layer_norm1.weight", "decoder.layers.2.pwff.layer_norm1.bias".
It seems a mismatch between Difnet_LRP architecture code and the pretrained model weights.
We are sorry for this, we have updated the code. Note that difnet_lrp
use a decoder as same as the base_lrp
.
when I use pretrained model DIFNet.pth, some errors reports:
RuntimeError: Error(s) in loading state_dict for Difnet_LRP: Missing key(s) in state_dict: "decoder.layers.0.pwff.layer_norm1.weight", "decoder.layers.0.pwff.layer_norm1.bias", "decoder.layers.1.pwff.layer_norm1.weight", "decoder.layers.1.pwff.layer_norm1.bias", "decoder.layers.2.pwff.layer_norm1.weight", "decoder.layers.2.pwff.layer_norm1.bias".
It seems a mismatch between Difnet_LRP architecture code and the pretrained model weights.
Our code has been updated, and the test results of our pre-trained difnet_lrp are as follows:
Thanks for replying
when I use pretrained model DIFNet.pth, some errors reports:
RuntimeError: Error(s) in loading state_dict for Difnet_LRP: Missing key(s) in state_dict: "decoder.layers.0.pwff.layer_norm1.weight", "decoder.layers.0.pwff.layer_norm1.bias", "decoder.layers.1.pwff.layer_norm1.weight", "decoder.layers.1.pwff.layer_norm1.bias", "decoder.layers.2.pwff.layer_norm1.weight", "decoder.layers.2.pwff.layer_norm1.bias".
It seems a mismatch between Difnet_LRP architecture code and the pretrained model weights.