Size of weights 32 does not match kernel shape: [32 1 3 3]

Maicus · ‎07-15-2019

Hello there,

I'm trying to convert an mxnet model to openvino.

The input shape of the images is 96x96 in grayscale.

When I'm trying to convert it with mo_mxnet.py then I get the following error:

[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:127 ]  --------------------
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:128 ]  Partial infer for conv_rd1_down_b
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:129 ]  Op: Const
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:140 ]  Inputs:
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:142 ]  Outputs:
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:34 ]  output[0]: shape = [32], value = [  3.9656727   3.1233442   7.8646393   4.28713     9.59885    -7.7935247
 -10.188561    3.451875 ...
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:127 ]  --------------------
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:128 ]  Partial infer for conv_rd1_down_w
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:129 ]  Op: Const
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:140 ]  Inputs:
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:142 ]  Outputs:
[ 2019-07-15 09:37:43,757 ] [ DEBUG ] [ infer:34 ]  output[0]: shape = [32  1  3  3], value = [[[[ 1.94546506e-01  1.07380378e+00  3.81941438e-01]
   [ 5.58629572e-01 -4.26461071e-01  1.97264...
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:127 ]  --------------------
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:128 ]  Partial infer for Input
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:129 ]  Op: Placeholder
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:140 ]  Inputs:
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:142 ]  Outputs:
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:34 ]  output[0]: shape = [ 1  1 96 96], value = <UNKNOWN>
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:127 ]  --------------------
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:128 ]  Partial infer for conv_rd1_down
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:129 ]  Op: Convolution
[ ERROR ]  Size of weights 32 does not match kernel shape: [32  1  3  3]
    Possible reason is wrong channel number in input shape

[ ERROR ]  Cannot infer shapes or values for node "conv_rd1_down".
[ ERROR ]  Cannot reshape weights to kernel shape
[ ERROR ]
[ ERROR ]  It can happen due to bug in custom shape infer function <function Convolution.infer at 0x00000201F0E8C378>.
[ ERROR ]  Or because the node inputs have incorrect values/shapes.
[ ERROR ]  Or because input shapes are incorrect (embedded to the model or passed via --input_shape).
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ infer:194 ]  Node "conv_rd1_down" attributes: {'symbol_dict': {'op': 'Convolution', 'name': 'conv_rd1_down', 'attrs': {'cudnn_off': '0', 'cudnn_tune': 'None', 'dilate': '(1,1)', 'kernel': '(3,3)', 'layout': 'None', 'no_bias': '0', 'num_filter': '32', 'num_group': '1', 'pad': '(1,1)', 'stride': '(2,2)', 'workspace': '1024'}, 'inputs': [[0, 0, 0], [1, 0, 0], [2, 0, 0]]}, '_in_ports': {0, 1, 2}, '_out_ports': {0}, 'kind': 'op', 'name': 'conv_rd1_down', 'type': 'Convolution', 'op': 'Convolution', 'infer': <function Convolution.infer at 0x00000201F0E8C378>, 'precision': 'FP32', 'multiplication_transparent': True, 'multiplication_transparent_ports': [(0, 0), (1, 0)], 'in_ports_count': 3, 'out_ports_count': 1, 'bias_addable': True, 'bias_term': False, 'pad': array([[0, 0],
       [0, 0],
       [1, 1],
       [1, 1]], dtype=int64), 'pad_spatial_shape': array([[1, 1],
       [1, 1]], dtype=int64), 'dilation': array([1, 1, 1, 1], dtype=int64), 'output_spatial_shape': None, 'output_shape': None, 'stride': array([1, 1, 2, 2], dtype=int64), 'group': 1, 'output': 32, 'kernel_spatial': array([3, 3], dtype=int64), 'input_feature_channel': 1, 'output_feature_channel': 0, 'kernel_spatial_idx': None, 'reshape_kernel': True, 'spatial_dims': None, 'channel_dims': array([1], dtype=int64), 'batch_dims': array([0], dtype=int64), 'layout': 'NCHW', 'dim_attrs': ['channel_dims', 'axis', 'batch_dims', 'spatial_dims'], 'shape_attrs': ['stride', 'shape', 'window', 'output_shape', 'pad'], 'IE': [('layer', [('id', <function Op.substitute_ie_attrs.<locals>.<lambda> at 0x00000201F323D158>), 'name', 'precision', 'type'], [('data', ['auto_pad', 'group', ('strides', <function Convolution.backend_attrs.<locals>.<lambda> at 0x00000201F323D1E0>), ('dilations', <function Convolution.backend_attrs.<locals>.<lambda> at 0x00000201F323D268>), ('kernel', <function Convolution.backend_attrs.<locals>.<lambda> at 0x00000201F323D2F0>), ('pads_begin', <function Convolution.backend_attrs.<locals>.<lambda> at 0x00000201F323D378>), ('pads_end', <function Convolution.backend_attrs.<locals>.<lambda> at 0x00000201F323D400>), 'output', 'pad_value', 'mode', 'input'], []), '@ports', '@consts'])], 'is_output_reachable': True, 'is_undead': False, 'is_const_producer': False, 'is_partial_inferred': False}
[ ERROR ]  Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.middle.PartialInfer.PartialInfer'>): Stopped shape/value propagation at "conv_rd1_down" node.
 For more information please refer to Model Optimizer FAQ (<INSTALL_DIR>/deployment_tools/documentation/docs/MO_FAQ.html), question #38.
[ 2019-07-15 09:37:43,772 ] [ DEBUG ] [ main:318 ]  Traceback (most recent call last):
  File "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\mo\middle\passes\infer.py", line 130, in partial_infer
    node.infer(node)
  File "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\mo\ops\convolution.py", line 146, in infer
    raise Error("Cannot reshape weights to kernel shape")
mo.utils.error.Error: Cannot reshape weights to kernel shape

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\mo\utils\class_registration.py", line 167, in apply_replacements
    replacer.find_and_replace_pattern(graph)
  File "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\extensions\middle\PartialInfer.py", line 31, in find_and_replace_pattern
    partial_infer(graph)
  File "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\mo\middle\passes\infer.py", line 196, in partial_infer
    refer_to_faq_msg(38)) from err
mo.utils.error.Error: Stopped shape/value propagation at "conv_rd1_down" node.
 For more information please refer to Model Optimizer FAQ (<INSTALL_DIR>/deployment_tools/documentation/docs/MO_FAQ.html), question #38.

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\mo\main.py", line 312, in main
    return driver(argv)
  File "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\mo\main.py", line 278, in driver
    ret_res = mo_mxnet.driver(argv, argv.input_model, model_name, argv.output_dir)
  File "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\mo\pipeline\mx.py", line 87, in driver
    class_registration.apply_replacements(graph, class_registration.ClassType.MIDDLE_REPLACER)
  File "C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\mo\utils\class_registration.py", line 184, in apply_replacements
    )) from err
mo.utils.error.Error: Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.middle.PartialInfer.PartialInfer'>): Stopped shape/value propagation at "conv_rd1_down" node.
 For more information please refer to Model Optimizer FAQ (<INSTALL_DIR>/deployment_tools/documentation/docs/MO_FAQ.html), question #38.

As I can see the converter thinks that the weight is 32, but in fact the convolution conv_rd1_down does have an input with the weight conv_rd1_down_w and the bias conv_rd1_down_b.

So the weights conv_rd1_down_w do have the right size of [32 1 3 3] so how does the model optimizer get the weight size of 32?

I attached the params file and the symbol for completeness.

The Command I called the optimizer:

python mo_mxnet.py --input_model hybridNet-0000.params --input_shape [1,1,96,96] --input Input --log_level=DEBUG

thanks in advance

Marcus

Kenneth_C_Intel · ‎07-17-2019

I have asked our developers to take a look at this

I will let you know what they say.

Thanks