search

chenzhi1992 / TensorRT-SSD

Use TensorRT API to implement Caffe-SSD， SSD（channel pruning）， Mobilenet-SSD
251 stars 84 forks source link

Do I just need to implement the softmax layer ? #14

Open ljq19950723 opened 6 years ago

ljq19950723 commented 6 years ago

Firstly,thank you for your code.And I compared the pluginImplement.cpp of your code with the samplePlugin.cpp of TensorRT's samplePlugin.Do I only need to implement the softmax layer ?And the mobilenet can work on TensorRT.

chenzhi1992 commented 6 years ago

You need to implement the depthwise layer， too.

ljq19950723 commented 6 years ago

And Do I need to implement the class SoftmaxPlugin's functions that inherited from the parent class only?Or any other functions?

chenzhi1992 commented 6 years ago

Yes， you need to implement the SoftmaxPlugin

xchani commented 6 years ago

I found the channels of output blobs from plugin reshape layer is actually the number of classes, hence off-the-shelf across-channel softmax layer can satisfy demand. @chenzhi1992

chenzhi1992 commented 6 years ago

Allright, you can verify that the result is correct.

twmht commented 6 years ago

@chenzhi1992

Do you have plan to release the code of softmax and depthwise layer?

linux-devil commented 6 years ago

@xchani Did off-the-shelf across-channel softmax worked for you ?

xchani commented 6 years ago

@linux-devil NO. Though off-the-shelf softmax and my own implemented softmax layer both operate across-channel, off-the-shelf softmax still produce wrong results. I cannot figure it out why it happends.

Optimus1072 commented 6 years ago

@xchani Did it work with your softmax implementation. Can you please share it and explain why the dimension of last layer is 12764?. Thanks :-)