liuyoude

It may because the path of dataset is set fasle, so the dataloader can't load batch data.

This function is same as the function of DCASE 2020 Task2 baseline. We directly extract the log-Mel spectrogram by [torchaudio](https://github.com/liuyoude/STgram_MFN/blob/7b5411c186656730586f52dfd74c226dc5fcca65/dataset.py#L9)

This seems to be caused by visdom visualization. You need run "python -m visdom.server" firstly. I will update this in README, thank you!

it only records the path of train files for dataset, you can see the details in run.py (preprocess() function).

> > it only records the path of train files for dataset, you can see the details in run.py (preprocess() function). > > > it only records the path of...

It's the number of IDs in development dataset and additional dataset. We train one model for all training data.

That's right.

请问这个问题怎么解决，我的也是没有发现v2rayL.service

It‘s also set as margin 0.7 and scale 30 in test.

Sorry, the margin and scale are the empirical values determined through experiments.