search

YuanGongND / vocalsound

Dataset and baseline code for the VocalSound dataset (ICASSP2022).
106 stars 10 forks source link

The avg precision is too low, my result is wrong? #2

Open joewale opened 2 years ago

joewale commented 2 years ago

I follow your instruction to run the project with the release vocalsound dataset, and I get the result as follows. The avg precision is too low, and my result is wrong?

image
YuanGongND commented 2 years ago

That is consistent with our exp (see the colab script result).

I am wondering if that is due to the save_every_step being too large in https://github.com/YuanGongND/vocalsound/blob/36fcc4be3f49655d1edf5a8bb065f0675d3a9818/src/utilities/stats.py#L52

Since the pipeline saves the raw output, probably you can take a look at the output of https://github.com/YuanGongND/vocalsound/blob/36fcc4be3f49655d1edf5a8bb065f0675d3a9818/src/traintest.py#L215

The accuracy score should be correct, and your result is consistent with that in our exp (you can easily reproduce using the colab scrpt at https://colab.research.google.com/github/YuanGongND/vocalsound/blob/main/colab/VocalSound.ipynb).

-Yuan