search

YangangCao / TRUNet

unofficial PyTorch implementation of 《REAL-TIME DENOISING AND DEREVERBERATION WTIH TINY RECURRENT U-NET》
93 stars 20 forks source link

T-GRU #4

Open shenbuguanni opened 2 years ago

shenbuguanni commented 2 years ago

T-GRU和F-GRU的输入shape好像是一样的，所以这里怎么提现这个T-GRU？

AmosCch commented 2 years ago

是的，按说这里应该再transpose一下，并且GRU的定义里应该要加上batch_first=True

YangangCao commented 2 years ago

T-GRU和F-GRU的输入shape好像是一样的，所以这里怎么提现这个T-GRU？

不一样，你打印一下看看

YangangCao commented 2 years ago

是的，按说这里应该再transpose一下，并且GRU的定义里应该要加上batch_first=True

不明白为什么要transpose，烦请赐教，确实要加batch_first=True，感谢提醒

AmosCch commented 2 years ago

是的，按说这里应该再transpose一下，并且GRU的定义里应该要加上batch_first=True

不明白为什么要transpose，烦请赐教，确实要加batch_first=True，感谢提醒 应该不需要transpose，我理解错了，我以为(1,4,257)中的4是时间帧，感谢提醒。

atabakp commented 2 years ago

The shape of TGRU's input(x9) is (Time, 16, 64). Since it should aggregate the information along the time-axis and batch_first=True in your implementation, therefore the input of TGRU should have "Time" should be the second dimension. Or set the batch_first=False for the TGRU.

amirpashamobinitehrani commented 1 year ago

@atabakp thanks for the useful info! Do you have any insights regarding the data input shape? seems to be a bit confusing. Referring to this thread here: https://github.com/YangangCao/TRUNet/issues/5#issue-1262713123