Message type “ditcaffe.LayerParameter” has no field named “permute_param”.

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX1
1

Hi experts,

I hit this issue when I infer MobileNetV1-SSD caffe model on my TX1. Seems TensorRT not support this layer.
Could anyone tell me how could I run MobileNetV1-SSD caffe model on my TX1? Any version of JetPack supports it?

Thanks in advance for any comments.

my TX1 info:
R28 (release), REVISION: 2.0, GCID: 10567845, BOARD: t210ref, EABI: aarch64, DATE: Fri Mar…
CUDA Version 9.0.252
#define CUDNN_MAJOR 7
nv-tensorrt-repo-ubuntu1604-ga-cuda9.0-trt3.0.4-20180208 1-1

2

Hi,

Could you try to comment the permute_param to see if works?

layer {
  name: ...
  type: ...
  bottom: ...
  top: ...
  #permute_param {
  #  ...
  #}
}

Thanks.

3

Hi microlj,

Have you tried the suggestion? Any result can be shared?

Thanks

4

Hi kayccc, AastaLLL,

sorry for delayed reply.

MobileNetV1-SSD now is working on my TX1. Since it not support SSD model directly, I used plugin layer to make them, permute, priorbox, detection_out, etc, worked. :)

Thanks.