michetonu / da-rnn_manoeuver_anticipation Goto Github PK

What is the reason of using Dense(2,activation='softmax') instead of Dense(1, activation='sigmoid')?

Is it related to the Gradient Reversal Layer? If so, can you explain?

Hey!

Thanks a lot for posting the code! I have one thing that I'm trying to understand. During training, we show the model a mix of both source domain data that have class labels, and target domain data that don't have them. How are missing class labels for target domain data handled? Do they have some distinct value (like -1)?

Hello! My main research direction is transfer fault diagnosis between machines! feature extractor is achieved by cnn with six con layers , six pooling layers and so on . so can you share the simple code of transfer learning in keras not da-rnn?
thanks!

Hi there,

Thanks for this implementation! I realise you're likely not working on this anymore, but I wanted to check if there is an error in the model setup, or if I have misunderstood something. Your action classification bit is as follows:

# Action classification - classify the manouever
        main_output = TimeDistributed(
            Dense(self.dense_units, activation='softmax', kernel_initializer=self.kernel_initializer),
            name='aux_classifier'
        )(aux_output)

From my understanding, the output of the model would be a softmax operation over a 128 dim vector. Based on your sketch, it seems like this should be a standard dense layer with tanh activation, followed by another dense layer with softmax over the 5 categories.

Hope my interpretation is correct - if I have misunderstood something, please let me know!