jungjee / asvspoof_pa Goto Github PK

Hi,

I am having issues running "g1-train_model.py" and "g2-train_train_model_CNN_GRU_wPretrn.py" as they are throwing the errors seen below respectively. Do you have any suggestions on how to access your models methods when wrapped in DataParallel? Would replacing model.METHOD_NAME -> model.module.METHOD_NAME be appropriate?

torch.nn.modules.module.ModuleAttributeError: 'DataParallel' object has no attribute 'compute_center_loss'
torch.nn.modules.module.ModuleAttributeError: 'DataParallel' object has no attribute 'forward_mode'

Thanks in advance!

Something worng when I am running g2-train_model_CNN_GRU_wPretrn.py,
No such file "spec_magnitude_2048_800_480/PA_T_0042447.npy"

Hello! Why does the program still report the error No module named 'comet_ml' after I have installed the 'comet_ml' library, is it because of a version problem? Can you post the version of the python library used in the experiment? Thanks!

Hi there,

I'm really interested in your amazing work, but I'm a little bit confused about the model architecture. Could you please help me figure out that?

What does the _make_layers() function actually do? It seems that the python implementation is slightly different from the model description in Table 1 mentioned in your paper, (namely variable first, # Resblocks and pooling strategy).

And could you also provide the phase , psd and raw implementation?

Thank you very much!

Best wishes,

Bill

Hello, could you share specCNN trained model?

Hi,
thank you for sharing comet logs! Could you check the comet link? It seems to me that the one you provided is somehow broken (there are 4 experiments under https://www.comet.ml/jungjee/public-comet-for-fithub/view/ but none of them is the one from this repository).

Thanks!

g1-train_model.py line 272
y_score.extend(out.cpu().numpy()[:,0]) #>>> (16, 64?)

Why take the first column instead of the one with the larger value？

I noticed that in your paper, only the EER of table5 does not add a percentage sign. Did you forget to add it?

Hi,

it seems that code for model does not exist, could you please upload those codes?

Thank you very much!