Hi,
I convert a IR TSM-ResNet(a video understanding model) from onnx, the xml file seems fine, whose input is
<net batch="1" name="tsm_res50_model_cpu_optimal" version="4">
<layers>
<layer id="0" name="0" precision="FP32" type="Input">
<output>
<port id="0">
<dim>16</dim>
<dim>3</dim>
<dim>224</dim>
<dim>224</dim>
</port>
</output>
</layer>
in which ,the "16" is not the batch size but the frames needed.
But when I load the model with python sample, the input shape become 1, 3, 244, 244! And the ie report an error
[ INFO ] Batch size is 1 [ INFO ] Loading model to the plugin Traceback (most recent call last): File "/MacUbuntuLocal/intel//computer_vision_sdk_2018.5.445/inference_engine/samples/python_samples/cxy_tsm.py", line 141, in <module> sys.exit(main() or 0) File "/MacUbuntuLocal/intel//computer_vision_sdk_2018.5.445/inference_engine/samples/python_samples/cxy_tsm.py", line 97, in main exec_net = plugin.load(network=net) File "ie_api.pyx", line 389, in openvino.inference_engine.ie_api.IEPlugin.load File "ie_api.pyx", line 400, in openvino.inference_engine.ie_api.IEPlugin.load RuntimeError: AssertionFailed: output->desc().dimsOrder() == inDesc.dimsOrder()
Did the loader change the 16 -> 1 to limit the batch size (which is not) to one? Any suggestions?
Thank you!
Link Copied
Any suggestion?
I've encountered similar question as below, it seems to be she shape problem too
Traceback (most recent call last): File "/MacUbuntuLocal/intel//computer_vision_sdk_2018.5.445/inference_engine/samples/python_samples/cxy_tsm.py", line 144, in <module> sys.exit(main() or 0) File "/MacUbuntuLocal/intel//computer_vision_sdk_2018.5.445/inference_engine/samples/python_samples/cxy_tsm.py", line 100, in main exec_net = plugin.load(network=net) File "ie_api.pyx", line 389, in openvino.inference_engine.ie_api.IEPlugin.load File "ie_api.pyx", line 400, in openvino.inference_engine.ie_api.IEPlugin.load RuntimeError: Incorrect blob sizes for node 384/Output_0/Data__const
Hi there. Where are you finding cxy_tsm.py ? I don't see it in the OpenVino installation. I have checked R4 and R5. Also please provide an exact URL to the ResNet model you are using.
Thanks,
Shubha
Shubha R. (Intel) wrote:Hi there. Where are you finding cxy_tsm.py ? I don't see it in the OpenVino installation. I have checked R4 and R5. Also please provide an exact URL to the ResNet model you are using.
Thanks,
Shubha
Thank you Suhbha, all the related files are listed here:
https://1drv.ms/f/s!Anu8COkYdog4hUKAdwUizLzJugl7
The cxy_tsm.py is a copy of classification_sample.py , with the input changed for testing.
This is a model with input of shape (8,3,224,224), "8" is the frame size, which means it needs eight image in one inference.
@ Shubha R.
Is there any progress for this problem?
Dear Soni, Neha,
Perhaps you are also the author of the below dldt github post ?
https://github.com/opencv/dldt/issues/227
In order to help you I would need a *.zip file containing your model as well as a short inference script which demonstrates the issue.
Thanks,
Shubha
For more complete information about compiler optimizations, see our Optimization Notice.