search

sharathadavanne / sed-crnn

Single and multichannel sound event detection using convolutional recurrent neural networks. DCASE 2017 real-life sound event detection winning method.
Other
184 stars 45 forks source link

How to test the trained model? #4

Closed JingpengSun closed 5 years ago

JingpengSun commented 5 years ago

Hi, I did find how to test one .h5 trained model by running sed.py?

sharathadavanne commented 5 years ago

Hi, I have only provided training scripts here, which learns from the training split and predicts on the validation split. If you want to test the trained model on a different dataset, then you will have to first load the trained model and test it with the corresponding features similar to this line

JingpengSun commented 5 years ago

Sorry for the late reply, I made it by following your answer, thank you! And I have an another question: should we must train the network from scratch on our own data?

sharathadavanne commented 5 years ago

Yes, you will have to train it yourself. Also note, this model is tuned for the DCASE 2017 SED dataset. If you are training on any other dataset, you might have to tune the model parameters (# layers, # filters/units/nodes per layer etc.)

JingpengSun commented 5 years ago

Oh, I see! I trained a model on our own dataset directly and got a terrible performance. I will explore the model further more and tune the model parameters. Many thanks~ I close it~