YuanGongND / ssast

Code for the AAAI 2022 paper "SSAST: Self-Supervised Audio Spectrogram Transformer".
BSD 3-Clause "New" or "Revised" License
365 stars 61 forks source link

Use Audioset20K for fine tuning #38

Open sunyulin0421 opened 2 months ago

sunyulin0421 commented 2 months ago

Hi,

Thank you for your nice work. I would like to ask why I get such a low mAP and AUC when using Audioset 20k dataset for fine-tuning in SSAST model?

For the dataset I took the preprocessing of resampling to 16kHz and the batch_size was set to 8 (12 in the original code) and the Epoch was also 25 rounds, but the mAP I got was only 0.004844 and the AUC was only 0.499981, which is unbelievable!

I wonder what your mAP and AUC are when fine-tuning with Audioset20K?

Looking forward to your reply, thanks!

result