How to fune-tune the pre-trained model on my custom dataset

[jmren168]

Thanks for the awesome work!

Is it possible to fine-tune the pre-trained model on my custom dataset? Thanks.

[SaoYear]

Hi, thanks for your interest.

Of course you can apply fine-tuning by your own dataset. Since the released codes of this work follows the DCASE task 4 baseline, you need to:

1. Re-organize your dataset to the DCASE task 4 (DESED) dataset structure, which includes: your audio waveforms + the meta file (.tsv file) following the meta file format of the DESED dataset.
2. Replace the waveform path and meta file path in the config files.
3. Following the 2-stage fine-tuning process in this work.

Hope it helps : )