audio-westlakeu / fn-ssl Goto Github PK

Dear author, thanks for your sharing of code and the state-of-the-art performance of Doa estimation. there are two problem confusing me:

1. in paper 2.1.2: if the target is from DP-IPD to localization classification, the activation will change from tanh to softmax, in readme, "set is_doa = True (Model.FN_SSL)", when I set the is_doa to True, seems the activation in Model.py is still tanh, the output shape is normal (B T 180), it can train normally, did I need to change the activation in Model.py to sofmax?
2. as said in the readme "the predgt2doa needs to be replaced synchronously", did I need to change it to predgt2DOA_cls?, when I change it to this, it has errors,

hope to see your answers.
best regards

I donn't know where to download the Noise92 dataset,could please give a link

I am using two RTX 3090 GPUs to run train.py following the provided guide.

I only used one FullNarrowBlock, but the training process took 5 hours to complete just one of the 15 epochs. The training process took too much time than I expected.

When I Checked the GPU utilization, it seemed there was a bottleneck somewhere in the code. I suspect the bottleneck might be in the data loading and processing part.

I am wondering if this is a normal occurrence.
If there is something wrong, could you give me an advice how to deal with it?

Hello, thanks for this algorithm. I would like to ask if it can locate 0 and 180 degrees. I don’t know why the positioning results never show 0 and 180 degrees.