NCS concat output reslut error

[idata]

My prototxt is like below:

I output single layer result(conv5_5_CPM_L1, conv5_5_CPM_L2, relu4_4_CPM_1) is all right(Use neural compute stick).However, I get the concat_stage2 output, the result is error. I get the concat_stage2 output use caffe on PC , the result right.

I doubt that is ncs Concat operation has something error..

Could someone help me ? ….….layer { name: "relu4_4_CPM_1" type: "ReLU" bottom: "conv4_4_CPM" top: "relu4_4_CPM_1"}….….layer { name: "conv5_5_CPM_L1" type: "Convolution" bottom: "conv5_4_CPM_L1" top: "conv5_5_CPM_L1" param { lr_mult: 1.0 decay_mult: 1 } param { lr_mult: 2.0 decay_mult: 0 } convolution_param { num_output: 38 pad: 0 kernel_size: 1 weight_filler { type: "gaussian" std: 0.01 } bias_filler { type: "constant" } }}layer { name: "conv5_5_CPM_L2" type: "Convolution" bottom: "conv5_4_CPM_L2" top: "conv5_5_CPM_L2" param { lr_mult: 1.0 decay_mult: 1 } param { lr_mult: 2.0 decay_mult: 0 } convolution_param { num_output: 19 pad: 0 kernel_size: 1 weight_filler { type: "gaussian" std: 0.01 } bias_filler { type: "constant" } }}layer { name: "concat_stage2" type: "Concat" bottom: "relu4_4_CPM_1" bottom: "conv5_5_CPM_L1" bottom: "conv5_5_CPM_L2" top: "concat_stage2" concat_param { axis: 1 }}

[idata]

:o

[idata]

I have the same problem. Have you fix it?

[idata]

@maqiao Have you tried using the latest NCSDK 2.05 and using the -ec option when compiling/checking your model?

[idata]

@Tome_at_Intel I have the following layers as my output layers :

layer {

name: "68point" type: "InnerProduct" bottom: "fc7" top: "68point" inner_product_param { num_output: 136 }}layer { name: "poselayer" type: "InnerProduct" bottom: "fc7" top: "poselayer" inner_product_param { num_output: 3 }}

I added a concat layer to the deploy.prototxt as below :

layer { name: "concat" bottom: "poselayer" bottom: "68point" top: "concat" type: "Concat" concat_param { axis:1 }}

When I try to profile it I get the following error :

ValueError: cannot reshape array of size 3 into shape (1,1,139)

When I add the -ec option I get the following error :

ValueError: cannot reshape array of size 136 into shape (1,1,139)

So only one final variable is added to the blob generated by the caffe parser and not both

[idata]

@karthik Is the Concat layer in your post above the last layer in your Caffe model? There is a known issue with NCSDK running Caffe and TensorFlow models that have the last layer being a Concat layer. Can you provide a link to your model so that I can reproduce the issue? Thanks.

[idata]

@Tome_at_Intel The model I use is from the following github repo : https://github.com/guozhongluo/head-pose-estimation-and-face-landmark/blob/master/model/deploy.prototxt

I've concatenated the last two layers as below :

layer {name: "concat"bottom: "poselayer"bottom: "68point"top: "concat"type: "Concat"concat_param {axis:1}}

Yup last layer is Concat, I was able to overcome this issue by adding a dummy reshape layer after concat.

[idata]

@karthik I'm glad to hear that you were able to find a workaround. Can you share the details of your reshape layer? I'm sure it would be useful to our community users. Thanks!

[idata]

The concat layer is as follows :

layer { name: "reshape" type: "Reshape" bottom: "concat" top: "reshape" reshape_param { shape { dim: 1 dim: 1 dim: 139 } } }

I'm very new to Caffe, It's a simple reshape layer I copied from the official site.

[idata]

@karthik

does the reshape layer in tensorflow can bypass the concat layer's bug ?