search

qiuqiangkong / audioset_tagging_cnn

MIT License
1.35k stars 255 forks source link

framewise output loss #14

Open manhlab opened 4 years ago

manhlab commented 4 years ago

Can you show me framewise_output loss?

qiuqiangkong commented 4 years ago

Hi manhlab, there is no framewise_output loss. All systems are trained with clipwise_output loss.

marcoyang1998 commented 2 years ago

So the sound event detection models are only trained with clipwise_output loss, but the forward function in those models is designed to generate framewise output? If so, can I modify a pretrained CNN_14 by just modifying its forward function, to make it perform framewise sound event detection without training from scratch again?

qiuqiangkong commented 2 years ago

Yes, you can apply CNN_14 framewisely.

On Mon, 4 Apr 2022 at 23:44, marcoyang1998 @.***> wrote:

So the sound event detection models are only trained with clipwise_output loss, but the forward function in those models is designed to generate framewise output? If so, can I modify a pretrained CNN_14 by just modifying its forward function, to make it perform framewise sound event detection without training from scratch again?

— Reply to this email directly, view it on GitHub https://github.com/qiuqiangkong/audioset_tagging_cnn/issues/14#issuecomment-1087719106, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADFXTSLJOBM6ND3UNBTSQOLVDME4ZANCNFSM4P56SKTQ . You are receiving this because you commented.Message ID: @.***>