Community
cancel
Showing results for 
Search instead for 
Did you mean: 
idata
Community Manager
352 Views

Support for Deconvolution

Hi all,

 

I am trying to make a simple network to learn the identity function, i.e., take an input image and give back the same image.

 

I am using the CIFAR10 dataset for training.

 

My architecture is as follows:

 

Input (32 x 32 x 3) -> Conv (8 filters, 5x5 filter size and 2,2 strides with same padding and ReLU activation) -> ConvTranspose (3 filters, 5x5 filter size and 2,2 strides with same padding and no activation).

 

I get the following error when I try to compile:

 

/usr/local/bin/ncsdk/Controllers/Parsers/TensorFlowParser/Convolution.py:46: SyntaxWarning: assertion is always true, perhaps remove parentheses? assert(False, "Layer type not supported by Convolution: " + obj.type) /usr/local/bin/ncsdk/Controllers/Parsers/Phases.py:322: SyntaxWarning: assertion is always true, perhaps remove parentheses? assert(len(pred) == 1, "Slice not supported to have >1 predecessors") mvNCCompile v02.00, Copyright @ Intel Corporation 2017 ****** Info: No Weights provided. inferred path: inference.data-00000-of-00001****** shape: (1, 32, 32, 3) [Error 5] Toolkit Error: Stage Details Not Supported: Wrong deconvolution output shape.

 

Any help will be greatly appreciated.

 

Here are my model files.

 

The command I am running is as follows:

 

mvNCCompile inference.meta -s 12 -in Input -on deconv1T/BiasAdd -o inference.graph
Tags (1)
0 Kudos
1 Reply
idata
Community Manager
59 Views

Using NCS API v2 fixed this issue.

Reply