Re: convert onnx to xml bin. it show me that Concat input shapes do not match.

zhong12 · ‎06-15-2020

lass TemporalShift(nn.Module):
    def __init__(self, net, n_segment=3, n_div=8, fold=8):
        super(TemporalShift, self).__init__()
        self.net = net
        self.n_segment = n_segment
        self.fold_div = n_div
        self.fold=fold
 
    def forward(self, x):
        x = shift(x, self.fold)
        out = self.net(x)
        return out
 
 
def shift(x, fold):
    n_segment = 10
    fold_div = 8
    nt, c, h, w = x.size()
 
    n_batch = 1
    x = x.view(n_batch, n_segment, c, h, w)
 
    left_side = torch.cat((x[:, 1:, :fold], torch.zeros(1, 1, fold, h, w)), dim=1)
    right_side = torch.cat((torch.zeros(1, 1, fold, h, w), x[:, :-1, fold: 2 * fold]), dim=1)
    out = torch.cat((left_side, right_side, x[:, :, 2 * fold:]), dim=2)
 
    return out.view(-1, c, h, w)

it is my code snippet about a custom layer above. the input parameter net was a normal convolution network. my openvino version is openvino_2020.2.120. when I try to convert . it show me that

(py36_env) zhx3@ubuntu:/opt/intel/openvino/deployment_tools/model_optimizer$ sudo python3 mo_onnx.py \
> --input_model tsm_resnet50.onnx \
> --input "input" \
> --input_shape [10,3,224,224] \
> --mean_values [0.4850,0.4560,0.4060] \
> --scale_values [0.2290,0.2240,0.2250]
Model Optimizer arguments:
Common parameters:
	- Path to the Input Model: 	/opt/intel/openvino_2020.2.120/deployment_tools/model_optimizer/tsm_resnet50.onnx
	- Path for generated IR: 	/opt/intel/openvino_2020.2.120/deployment_tools/model_optimizer/.
	- IR output name: 	tsm_resnet50
	- Log level: 	ERROR
	- Batch: 	Not specified, inherited from the model
	- Input layers: 	input
	- Output layers: 	Not specified, inherited from the model
	- Input shapes: 	[10,3,224,224]
	- Mean values: 	[0.4850,0.4560,0.4060]
	- Scale values: 	[0.2290,0.2240,0.2250]
	- Scale factor: 	Not specified
	- Precision of IR: 	FP32
	- Enable fusing: 	True
	- Enable grouped convolutions fusing: 	True
	- Move mean values to preprocess section: 	False
	- Reverse input channels: 	False
ONNX specific parameters:
Model Optimizer version: 	2020.2.0-60-g0bc66e26ff
[ ERROR ]  Concat input shapes do not match
[ ERROR ]  Shape is not defined for output 0 of "390".
[ ERROR ]  Cannot infer shapes or values for node "390".
[ ERROR ]  Not all output shapes were inferred or fully defined for node "390". 
 For more information please refer to Model Optimizer FAQ (https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_Model_Optimizer_FAQ.html), question #40. 
[ ERROR ]  
[ ERROR ]  It can happen due to bug in custom shape infer function <function concat_infer at 0x7f8b2040ad08>.
[ 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).
[ ERROR ]  Run Model Optimizer with --log_level=DEBUG for more information.
[ ERROR ]  Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.middle.PartialInfer.PartialInfer'>): Stopped shape/value propagation at "390" node. 
 For more information please refer to Model Optimizer FAQ (https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_Model_Optimizer_FAQ.html), question #38.

[ 2020-06-15 17:39:40,433 ] [ DEBUG ] [ infer:144 ]  Outputs:
[ 2020-06-15 17:39:40,433 ] [ DEBUG ] [ infer:31 ]  output[0]: shape = [ 1 10  8 56 56], value = <UNKNOWN>
[ 2020-06-15 17:39:40,434 ] [ DEBUG ] [ infer:128 ]  --------------------
[ 2020-06-15 17:39:40,434 ] [ DEBUG ] [ infer:129 ]  Partial infer for 390
[ 2020-06-15 17:39:40,434 ] [ DEBUG ] [ infer:130 ]  Op: Concat
[ 2020-06-15 17:39:40,434 ] [ DEBUG ] [ infer:131 ]  Inputs:
[ 2020-06-15 17:39:40,434 ] [ DEBUG ] [ infer:31 ]  input[0]: shape = [ 1 10  8 56 56], value = <UNKNOWN>
[ 2020-06-15 17:39:40,434 ] [ DEBUG ] [ infer:31 ]  input[1]: shape = [ 1  2  8 56 56], value = <UNKNOWN>
[ 2020-06-15 17:39:40,434 ] [ DEBUG ] [ infer:31 ]  input[2]: shape = [ 1 10 48 56 56], value = <UNKNOWN>
[ ERROR ]  Concat input shapes do not match
[ 2020-06-15 17:39:40,435 ] [ DEBUG ] [ infer:144 ]  Outputs:
[ 2020-06-15 17:39:40,435 ] [ DEBUG ] [ infer:31 ]  output[0]: shape = <UNKNOWN>, value = <UNKNOWN>
[ ERROR ]  Shape is not defined for output 0 of "390".
[ ERROR ]  Cannot infer shapes or values for node "390".
[ ERROR ]  Not all output shapes were inferred or fully defined for node "390". 
 For more information please refer to Model Optimizer FAQ (https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_Model_Optimizer_FAQ.html), question #40.

above is my part log file.

zhong12 · ‎06-15-2020

I have solve the problem. because the openvino may not know the python idx -1 such as the x[:, :- 1, fold: 2 * fold] the :- 1 slice will get wrong shape

JAVIERJOSE_A_Intel · ‎06-15-2020

Hi zhong12,

We are glad to hear you solved your issue. Thanks for sharing your solution with the community!

Regards,

Javier A.