Errors when using tensorRT2 to accelerate SSD(caffe)

Autonomous Machines Jetson & Embedded Systems Jetson TX1
1

I want to accelerate SSD by adding other plugin layers into the sampleFasterRCNN. But I meet some errors at layer ‘mbox_loc’ and ‘mbox_conf’ when I put the plugins together:
“Plugin layer output count is not equal to caffe output count”.
What does that means? And could you please send me the source code of the tensorRT2.1?

2

Hi,

Please make sure the output blob number is equal.

  1. Plugin: int getNbOutputs()
  2. Caffe: definition of prototxt file, top: xxx

TensorRT doesn’t an opensource library. But you can find some sample code here:

Thanks.

3

Hi,

Here is the code:

int getNbOutputs() const override
{
std::cout << “getNbOutputs” << std::endl;
return 1;
}

layer {
name: “mbox_loc”
type: “IPlugin”
bottom: “conv4_3_norm_mbox_loc_flat”
bottom: “fc7_mbox_loc_flat”
bottom: “conv6_2_mbox_loc_flat”
bottom: “conv7_2_mbox_loc_flat”
bottom: “conv8_2_mbox_loc_flat”
bottom: “conv9_2_mbox_loc_flat”
top: “mbox_loc”
}

But there is no output string “getNbOutputs”!

Thanks.

4

Hi,

Please remember to add your plugin to plugin factory:
[url]Face-Recognition/pluginImplement.cpp at master · AastaNV/Face-Recognition · GitHub

Thanks.

5

Hi,

I figure it out! That is because I forget to call function mConcatPlugin.get(). So sad.

Thanks.

6

Hi @2013302477 are you able to share the source, I have been stuck with the problem of optimizing SSD with tensorRT the recent weeks.