Dnn pytorch Text Detection model to ONNX format

Python
1

I need to know how to convert a trained model based on totaltext_resnet50 model to Onnx.
I used for the training the GitHub - MhLiao/DB: A PyTorch implementation of "Real-time Scene Text Detection with Differentiable Binarization". repo.
My pytorch version : 1.8.0+cu111 .
The exception message I received : ONNX export failed: Couldn’t export Python operator Scatter Defined at: C:\Program Files (x86)\Microsoft Visual Studio\Shared …

The opencv/opencv github repo suggests to do exactly what I want. See it here : opencv/dnn_text_spotting.markdown at master · opencv/opencv · GitHub
, and look for the line : “You can train your own model with more data, and convert it into ONNX format. We encourage you to add new algorithms to these APIs.”

I will appreciate it a lot if this documentation could be extended with a link or code sample of how to do the conversion process to Onnx format . I believe that all the text detection pre-trained models were obtained that way( see the dnn_text_spotting/dnn_text_spotting.markdown link) .

I would prefer python code that solves this problem or maybe an Onnxruntime way.
Thanks.

2

please share your conversion code here

3

//// CONVERSATION CODE
//// BASICALLY : I LOAD A MODEL IN THE SAME WAY THAT THE demo.py DOES
//// IN THE https://github.com/MhLiao/DB.
//// THEN WHEN THE inference METHOD IS RUNNING I ADD ALL THE CODE VARIANTS THAT I COULD IMAGINE
//// TO CONVERT THE MODEL TO AN ONNX FORMAT AND THEY DON’T WORK.
//// NOTE : THE demo.py WORKS VERY WELL IF I USE THE ORIGINAL CODE WITH NO INTENTION TO CONVERT TO ONNX FORMAT .

//// HERE EXAMPLES :
//// FOR EACH EXAMPLE I TRIED BEFORE AND AFTER THE “pred = model.forward(batch, training=False)” LINE 137
//// INSIDE THE inference METHOD AFTER THE model.eval() LINE 130

//// EXAMPLE 1 : USING THE https://michhar.github.io/convert-pytorch-onnx/

// dummy_input = torch.randn(sample_batch_size, channel, height, width) //
dummy_input = torch.randn(img.shape[0],img.shape[1],img.shape[2],img.shape[3])
torch.onnx.export(model, dummy_input, “onnx_model_name.onnx”)

//// EXAMPLE 2 : USING THE https://deci.ai/resources/blog/how-to-convert-a-pytorch-model-to-onnx/

// dummy_input = torch.randn(1, 3, 224, 224) //
dummy_input = torch.randn(img.shape[0],img.shape[1],img.shape[2],img.shape[3])
input_names = [ “actual_input” ]
output_names = [ “output” ]
torch.onnx.export(model,
dummy_input,
“resnet50.onnx”,
verbose=False,
input_names=input_names,
output_names=output_names,
export_params=True,
)

//// EXAMPLE 3 : USING THE https://pytorch.org/tutorials/advanced/super_resolution_with_onnxruntime.html

    torch_model = model
        x = torch.randn(img.shape[0],img.shape[1],img.shape[2],img.shape[3])
        torch_out = torch_model(x)
        torch.onnx.export(torch_model,               # model being run
              x,                         # model input (or a tuple for multiple inputs)
              "super_resolution.onnx",   # where to save the model (can be a file or file-like object)
              export_params=True,        # store the trained parameter weights inside the model file
              opset_version=10,          # the ONNX version to export the model to
              do_constant_folding=True,  # whether to execute constant folding for optimization
              input_names = ['input'],   # the model's input names
              output_names = ['output'], # the model's output names
              dynamic_axes={'input' : {0 : 'batch_size'},    # variable length axes
                            'output' : {0 : 'batch_size'}})
4

Scatter layer is deprecated, but available in opsets 9, 11

so try changing your opset_version (you got 10 there)

apart from that, only thing im missing from the export code is a line like:

model.eval()

can you give us the complete error msg ? (wtf is it doing with visual studio there ?)

5

Hi Berak,

Thanks a lot for your answer !

Right now, I am working on your suggestions .
By the way, model.eval() is executed in the inference method before I added my code so I cannot explain what happens but I will try with the Scatter layer deprecated information .

I added the error message :

ONNX export failed: Couldn't export Python operator Scatter Defined at: C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\parallel\scatter_gather.py(19): scatter_map C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\parallel\scatter_gather.py(23): scatter_map C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\parallel\scatter_gather.py(36): scatter C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\parallel\scatter_gather.py(44): scatter_kwargs C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\parallel\data_parallel.py(174): scatter C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\parallel\data_parallel.py(157): forward C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\module.py(860): _slow_forward C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\module.py(887): _call_impl D:\Development\work now\emgu_newFeatures\DB\structure\model.py(56): forward C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\module.py(860): _slow_forward C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\module.py(887): _call_impl C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\jit\_trace.py(116): wrapper C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\jit\_trace.py(130): forward C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\module.py(889): _call_impl C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\jit\_trace.py(1139): _get_trace_graph C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\onnx\utils.py(377): _trace_and_get_graph_from_model C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\onnx\utils.py(417): _create_jit_graph C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\onnx\utils.py(456): _model_to_graph C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\onnx\utils.py(698): _export C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\onnx\utils.py(94): export C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\onnx\__init__.py(276): export D:\Development\work now\emgu_newFeatures\DB\demo.py(159): inference D:\Development\work now\emgu_newFeatures\DB\demo.py(50): main D:\Development\work now\emgu_newFeatures\DB\demo.py(178): <module> c:\program files (x86)\microsoft visual studio\2017\professional\common7\ide\extensions\microsoft\python\core\Packages\ptvsd\_vendored\pydevd\_pydev_imps\_pydev_execfile.py(25): execfile c:\program files (x86)\microsoft visual studio\2017\professional\common7\ide\extensions\microsoft\python\core\Packages\ptvsd\_vendored\pydevd\pydevd.py(1106): _exec c:\program files (x86)\microsoft visual studio\2017\professional\common7\ide\extensions\microsoft\python\core\Packages\ptvsd\_vendored\pydevd\pydevd.py(1099): run c:\program files (x86)\microsoft visual studio\2017\professional\common7\ide\extensions\microsoft\python\core\Packages\ptvsd\_vendored\pydevd\pydevd.py(1752): main c:\program files (x86)\microsoft visual studio\2017\professional\common7\ide\extensions\microsoft\python\core\Packages\ptvsd\_local.py(125): _run c:\program files (x86)\microsoft visual studio\2017\professional\common7\ide\extensions\microsoft\python\core\Packages\ptvsd\_local.py(64): run_file c:\program files (x86)\microsoft visual studio\2017\professional\common7\ide\extensions\microsoft\python\core\Packages\ptvsd\debugger.py(37): debug c:\program files (x86)\microsoft visual studio\2017\professional\common7\ide\extensions\microsoft\python\core\ptvsd_launcher.py(119): <module> Graph we tried to export: graph(%input : Float(1, 3, 736, 960, strides=[2119680, 706560, 960, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.0.conv2_offset.weight : Float(27, 128, 3, 3, strides=[1152, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.0.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.0.conv2.weight : Float(128, 128, 3, 3, strides=[1152, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.0.bn2.weight : Float(128, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.0.bn2.bias : Float(128, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.0.bn2.running_mean : Float(128, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer2.0.bn2.running_var : Float(128, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer2.1.conv2_offset.weight : Float(27, 128, 3, 3, strides=[1152, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.1.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.1.conv2.weight : Float(128, 128, 3, 3, strides=[1152, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.1.bn2.weight : Float(128, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.1.bn2.bias : Float(128, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.1.bn2.running_mean : Float(128, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer2.1.bn2.running_var : Float(128, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer2.2.conv2_offset.weight : Float(27, 128, 3, 3, strides=[1152, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.2.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.2.conv2.weight : Float(128, 128, 3, 3, strides=[1152, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.2.bn2.weight : Float(128, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.2.bn2.bias : Float(128, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.2.bn2.running_mean : Float(128, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer2.2.bn2.running_var : Float(128, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer2.3.conv2_offset.weight : Float(27, 128, 3, 3, strides=[1152, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.3.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.3.conv2.weight : Float(128, 128, 3, 3, strides=[1152, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.3.bn2.weight : Float(128, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.3.bn2.bias : Float(128, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer2.3.bn2.running_mean : Float(128, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer2.3.bn2.running_var : Float(128, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.0.conv2_offset.weight : Float(27, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.0.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.0.conv2.weight : Float(256, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.0.bn2.weight : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.0.bn2.bias : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.0.bn2.running_mean : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.0.bn2.running_var : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.1.conv2_offset.weight : Float(27, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.1.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.1.conv2.weight : Float(256, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.1.bn2.weight : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.1.bn2.bias : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.1.bn2.running_mean : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.1.bn2.running_var : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.2.conv2_offset.weight : Float(27, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.2.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.2.conv2.weight : Float(256, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.2.bn2.weight : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.2.bn2.bias : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.2.bn2.running_mean : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.2.bn2.running_var : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.3.conv2_offset.weight : Float(27, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.3.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.3.conv2.weight : Float(256, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.3.bn2.weight : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.3.bn2.bias : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.3.bn2.running_mean : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.3.bn2.running_var : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.4.conv2_offset.weight : Float(27, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.4.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.4.conv2.weight : Float(256, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.4.bn2.weight : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.4.bn2.bias : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.4.bn2.running_mean : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.4.bn2.running_var : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.5.conv2_offset.weight : Float(27, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.5.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.5.conv2.weight : Float(256, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.5.bn2.weight : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.5.bn2.bias : Float(256, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer3.5.bn2.running_mean : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer3.5.bn2.running_var : Float(256, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer4.0.conv2_offset.weight : Float(27, 512, 3, 3, strides=[4608, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.0.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.0.conv2.weight : Float(512, 512, 3, 3, strides=[4608, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.0.bn2.weight : Float(512, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.0.bn2.bias : Float(512, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.0.bn2.running_mean : Float(512, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer4.0.bn2.running_var : Float(512, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer4.1.conv2_offset.weight : Float(27, 512, 3, 3, strides=[4608, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.1.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.1.conv2.weight : Float(512, 512, 3, 3, strides=[4608, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.1.bn2.weight : Float(512, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.1.bn2.bias : Float(512, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.1.bn2.running_mean : Float(512, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer4.1.bn2.running_var : Float(512, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer4.2.conv2_offset.weight : Float(27, 512, 3, 3, strides=[4608, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.2.conv2_offset.bias : Float(27, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.2.conv2.weight : Float(512, 512, 3, 3, strides=[4608, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.2.bn2.weight : Float(512, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.2.bn2.bias : Float(512, strides=[1], requires_grad=1, device=cuda:0), %model.module.backbone.layer4.2.bn2.running_mean : Float(512, strides=[1], requires_grad=0, device=cuda:0), %model.module.backbone.layer4.2.bn2.running_var : Float(512, strides=[1], requires_grad=0, device=cuda:0), %model.module.decoder.in5.weight : Float(256, 2048, 1, 1, strides=[2048, 1, 1, 1], requires_grad=1, device=cuda:0), %model.module.decoder.in4.weight : Float(256, 1024, 1, 1, strides=[1024, 1, 1, 1], requires_grad=1, device=cuda:0), %model.module.decoder.in3.weight : Float(256, 512, 1, 1, strides=[512, 1, 1, 1], requires_grad=1, device=cuda:0), %model.module.decoder.in2.weight : Float(256, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=1, device=cuda:0), %model.module.decoder.out5.0.weight : Float(64, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.decoder.out4.0.weight : Float(64, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.decoder.out3.0.weight : Float(64, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.decoder.out2.weight : Float(64, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=1, device=cuda:0), %model.module.decoder.binarize.3.weight : Float(64, 64, 2, 2, strides=[256, 4, 2, 1], requires_grad=1, device=cuda:0), %model.module.decoder.binarize.3.bias : Float(64, strides=[1], requires_grad=1, device=cuda:0), %model.module.decoder.binarize.4.weight : Float(64, strides=[1], requires_grad=1, device=cuda:0), %model.module.decoder.binarize.4.bias : Float(64, strides=[1], requires_grad=1, device=cuda:0), %model.module.decoder.binarize.4.running_mean : Float(64, strides=[1], requires_grad=0, device=cuda:0), %model.module.decoder.binarize.4.running_var : Float(64, strides=[1], requires_grad=0, device=cuda:0), %model.module.decoder.binarize.6.weight : Float(64, 1, 2, 2, strides=[4, 4, 2, 1], requires_grad=1, device=cuda:0), %model.module.decoder.binarize.6.bias : Float(1, strides=[1], requires_grad=1, device=cuda:0), %659 : Float(64, 3, 7, 7, strides=[147, 49, 7, 1], requires_grad=0, device=cuda:0), %660 : Float(64, strides=[1], requires_grad=0, device=cuda:0), %662 : Float(64, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0), %663 : Float(64, strides=[1], requires_grad=0, device=cuda:0), %665 : Float(64, 64, 3, 3, strides=[576, 9, 3, 1], requires_grad=0, device=cuda:0), %666 : Float(64, strides=[1], requires_grad=0, device=cuda:0), %668 : Float(256, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0), %669 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %671 : Float(256, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0), %672 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %674 : Float(64, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %675 : Float(64, strides=[1], requires_grad=0, device=cuda:0), %677 : Float(64, 64, 3, 3, strides=[576, 9, 3, 1], requires_grad=0, device=cuda:0), %678 : Float(64, strides=[1], requires_grad=0, device=cuda:0), %680 : Float(256, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0), %681 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %683 : Float(64, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %684 : Float(64, strides=[1], requires_grad=0, device=cuda:0), %686 : Float(64, 64, 3, 3, strides=[576, 9, 3, 1], requires_grad=0, device=cuda:0), %687 : Float(64, strides=[1], requires_grad=0, device=cuda:0), %689 : Float(256, 64, 1, 1, strides=[64, 1, 1, 1], requires_grad=0, device=cuda:0), %690 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %692 : Float(128, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %693 : Float(128, strides=[1], requires_grad=0, device=cuda:0), %695 : Float(512, 128, 1, 1, strides=[128, 1, 1, 1], requires_grad=0, device=cuda:0), %696 : Float(512, strides=[1], requires_grad=0, device=cuda:0), %698 : Float(512, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %699 : Float(512, strides=[1], requires_grad=0, device=cuda:0), %701 : Float(128, 512, 1, 1, strides=[512, 1, 1, 1], requires_grad=0, device=cuda:0), %702 : Float(128, strides=[1], requires_grad=0, device=cuda:0), %704 : Float(512, 128, 1, 1, strides=[128, 1, 1, 1], requires_grad=0, device=cuda:0), %705 : Float(512, strides=[1], requires_grad=0, device=cuda:0), %707 : Float(128, 512, 1, 1, strides=[512, 1, 1, 1], requires_grad=0, device=cuda:0), %708 : Float(128, strides=[1], requires_grad=0, device=cuda:0), %710 : Float(512, 128, 1, 1, strides=[128, 1, 1, 1], requires_grad=0, device=cuda:0), %711 : Float(512, strides=[1], requires_grad=0, device=cuda:0), %713 : Float(128, 512, 1, 1, strides=[512, 1, 1, 1], requires_grad=0, device=cuda:0), %714 : Float(128, strides=[1], requires_grad=0, device=cuda:0), %716 : Float(512, 128, 1, 1, strides=[128, 1, 1, 1], requires_grad=0, device=cuda:0), %717 : Float(512, strides=[1], requires_grad=0, device=cuda:0), %719 : Float(256, 512, 1, 1, strides=[512, 1, 1, 1], requires_grad=0, device=cuda:0), %720 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %722 : Float(1024, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %723 : Float(1024, strides=[1], requires_grad=0, device=cuda:0), %725 : Float(1024, 512, 1, 1, strides=[512, 1, 1, 1], requires_grad=0, device=cuda:0), %726 : Float(1024, strides=[1], requires_grad=0, device=cuda:0), %728 : Float(256, 1024, 1, 1, strides=[1024, 1, 1, 1], requires_grad=0, device=cuda:0), %729 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %731 : Float(1024, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %732 : Float(1024, strides=[1], requires_grad=0, device=cuda:0), %734 : Float(256, 1024, 1, 1, strides=[1024, 1, 1, 1], requires_grad=0, device=cuda:0), %735 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %737 : Float(1024, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %738 : Float(1024, strides=[1], requires_grad=0, device=cuda:0), %740 : Float(256, 1024, 1, 1, strides=[1024, 1, 1, 1], requires_grad=0, device=cuda:0), %741 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %743 : Float(1024, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %744 : Float(1024, strides=[1], requires_grad=0, device=cuda:0), %746 : Float(256, 1024, 1, 1, strides=[1024, 1, 1, 1], requires_grad=0, device=cuda:0), %747 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %749 : Float(1024, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %750 : Float(1024, strides=[1], requires_grad=0, device=cuda:0), %752 : Float(256, 1024, 1, 1, strides=[1024, 1, 1, 1], requires_grad=0, device=cuda:0), %753 : Float(256, strides=[1], requires_grad=0, device=cuda:0), %755 : Float(1024, 256, 1, 1, strides=[256, 1, 1, 1], requires_grad=0, device=cuda:0), %756 : Float(1024, strides=[1], requires_grad=0, device=cuda:0), %758 : Float(512, 1024, 1, 1, strides=[1024, 1, 1, 1], requires_grad=0, device=cuda:0), %759 : Float(512, strides=[1], requires_grad=0, device=cuda:0), %761 : Float(2048, 512, 1, 1, strides=[512, 1, 1, 1], requires_grad=0, device=cuda:0), %762 : Float(2048, strides=[1], requires_grad=0, device=cuda:0), %764 : Float(2048, 1024, 1, 1, strides=[1024, 1, 1, 1], requires_grad=0, device=cuda:0), %765 : Float(2048, strides=[1], requires_grad=0, device=cuda:0), %767 : Float(512, 2048, 1, 1, strides=[2048, 1, 1, 1], requires_grad=0, device=cuda:0), %768 : Float(512, strides=[1], requires_grad=0, device=cuda:0), %770 : Float(2048, 512, 1, 1, strides=[512, 1, 1, 1], requires_grad=0, device=cuda:0), %771 : Float(2048, strides=[1], requires_grad=0, device=cuda:0), %773 : Float(512, 2048, 1, 1, strides=[2048, 1, 1, 1], requires_grad=0, device=cuda:0), %774 : Float(512, strides=[1], requires_grad=0, device=cuda:0), %776 : Float(2048, 512, 1, 1, strides=[512, 1, 1, 1], requires_grad=0, device=cuda:0), %777 : Float(2048, strides=[1], requires_grad=0, device=cuda:0), %779 : Float(64, 256, 3, 3, strides=[2304, 9, 3, 1], requires_grad=0, device=cuda:0), %780 : Float(64, strides=[1], requires_grad=0, device=cuda:0), %781 : Float(4, strides=[1], requires_grad=0, device=cuda:0), %782 : Float(4, strides=[1], requires_grad=0, device=cuda:0), %783 : Float(4, strides=[1], requires_grad=0, device=cuda:0), %784 : Float(4, strides=[1], requires_grad=0, device=cuda:0), %785 : Float(4, strides=[1], requires_grad=0, device=cuda:0), %786 : Float(4, strides=[1], requires_grad=0, device=cuda:0)): %387 : Float(1, 3, 736, 960, strides=[2119680, 706560, 960, 1], requires_grad=1, device=cuda:0) = onnx::Cast[to=1](%input) # D:\Development\work now\emgu_newFeatures\DB\structure\model.py:54:0 %388 : Float(1, 3, 736, 960, strides=[2119680, 706560, 960, 1], requires_grad=1, device=cuda:0) = onnx::Cast[to=1](%387) # D:\Development\work now\emgu_newFeatures\DB\structure\model.py:55:0 %389 : Tensor = ^Scatter([0], None, 0)(%388) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\parallel\scatter_gather.py:19:0 %658 : Float(1, 64, 368, 480, strides=[11304960, 176640, 480, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[7, 7], pads=[3, 3, 3, 3], strides=[2, 2]](%389, %659, %660) %392 : Float(1, 64, 368, 480, strides=[11304960, 176640, 480, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%658) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %393 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::MaxPool[kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%392) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:659:0 %661 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%393, %662, %663) %396 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%661) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %664 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%396, %665, %666) %399 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%664) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %667 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%399, %668, %669) %670 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%393, %671, %672) %404 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Add(%667, %670) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0 %405 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%404) %673 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%405, %674, %675) %408 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%673) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %676 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%408, %677, %678) %411 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%676) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %679 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%411, %680, %681) %414 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Add(%679, %405)

6

The message was very long and I had to divide it. This is the 2nd part :

# D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0 %415 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%414) %682 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%415, %683, %684) %418 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%682) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %685 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%418, %686, %687) %421 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%685) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %688 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%421, %689, %690) %424 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Add(%688, %415) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0 %425 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%424) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %691 : Float(1, 128, 184, 240, strides=[5652480, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%425, %692, %693) %428 : Float(1, 128, 184, 240, strides=[5652480, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%691) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %429 : Float(1, 27, 184, 240, strides=[1192320, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%428, %model.module.backbone.layer2.0.conv2_offset.weight, %model.module.backbone.layer2.0.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %430 : Float(1, 18, 184, 240, strides=[1192320, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%429) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0 %431 : Float(1, *, 184, 240, strides=[1192320, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%429) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %432 : Float(1, *, 184, 240, strides=[397440, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%431) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %input.37 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 2, 1, 1, 1, 1)(%428, %430, %432, %model.module.backbone.layer2.0.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0 %434 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.37, %model.module.backbone.layer2.0.bn2.weight, %model.module.backbone.layer2.0.bn2.bias, %model.module.backbone.layer2.0.bn2.running_mean, %model.module.backbone.layer2.0.bn2.running_var)# C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0 %435 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%434) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %694 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%435, %695, %696) %697 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[2, 2]](%425, %698, %699) %440 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Add(%694, %697) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0 %441 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%440) %700 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%441, %701, %702) %444 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%700) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %445 : Float(1, 27, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%444, %model.module.backbone.layer2.1.conv2_offset.weight, %model.module.backbone.layer2.1.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %446 : Float(1, 18, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%445) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0 %447 : Float(1, *, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%445) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %448 : Float(1, *, 92, 120, strides=[99360, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%447) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %input.47 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%444, %446, %448, %model.module.backbone.layer2.1.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0 %450 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.47, %model.module.backbone.layer2.1.bn2.weight, %model.module.backbone.layer2.1.bn2.bias, %model.module.backbone.layer2.1.bn2.running_mean, %model.module.backbone.layer2.1.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0 %451 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%450) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %703 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%451, %704, %705) %454 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Add(%703, %441) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0 %455 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%454) %706 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%455, %707, %708) %458 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%706) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %459 : Float(1, 27, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%458, %model.module.backbone.layer2.2.conv2_offset.weight, %model.module.backbone.layer2.2.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %460 : Float(1, 18, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%459) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0 %461 : Float(1, *, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%459) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %462 : Float(1, *, 92, 120, strides=[99360, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%461) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %input.56 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%458, %460, %462, %model.module.backbone.layer2.2.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0 %464 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.56, %model.module.backbone.layer2.2.bn2.weight, %model.module.backbone.layer2.2.bn2.bias, %model.module.backbone.layer2.2.bn2.running_mean, %model.module.backbone.layer2.2.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0 %465 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%464) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %709 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%465, %710, %711) %468 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Add(%709, %455) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0 %469 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%468) %712 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%469, %713, %714) %472 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%712) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %473 : Float(1, 27, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%472, %model.module.backbone.layer2.3.conv2_offset.weight, %model.module.backbone.layer2.3.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %474 : Float(1, 18, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%473) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0 %475 : Float(1, *, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%473) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %476 : Float(1, *, 92, 120, strides=[99360, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%475) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %input.65 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%472, %474, %476, %model.module.backbone.layer2.3.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0 %478 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.65, %model.module.backbone.layer2.3.bn2.weight, %model.module.backbone.layer2.3.bn2.bias, %model.module.backbone.layer2.3.bn2.running_mean, %model.module.backbone.layer2.3.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0 %479 : Float(1, 128, 92, 120, strides=[1413120, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%478) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %715 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%479, %716, %717) %482 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Add(%715, %469) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0 %483 : Float(1, 512, 92, 120, strides=[5652480, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%482) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %718 : Float(1, 256, 92, 120, strides=[2826240, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%483, %719, %720) %486 : Float(1, 256, 92, 120, strides=[2826240, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%718) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %487 : Float(1, 27, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%486, %model.module.backbone.layer3.0.conv2_offset.weight, %model.module.backbone.layer3.0.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %488 : Float(1, 18, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%487) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0 %489 : Float(1, *, 92, 120, strides=[298080, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%487) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %490 : Float(1, *, 92, 120, strides=[99360, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%489) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %input.74 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 2, 1, 1, 1, 1)(%486, %488, %490, %model.module.backbone.layer3.0.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0 %492 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.74, %model.module.backbone.layer3.0.bn2.weight, %model.module.backbone.layer3.0.bn2.bias, %model.module.backbone.layer3.0.bn2.running_mean, %model.module.backbone.layer3.0.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0 %493 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%492) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %721 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%493, %722, %723) %724 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[2, 2]](%483, %725, %726) %498 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Add(%721, %724) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0 %499 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%498) %727 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%499, %728, %729) %502 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%727) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %503 : Float(1, 27, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%502, %model.module.backbone.layer3.1.conv2_offset.weight, %model.module.backbone.layer3.1.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %504 : Float(1, 18, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%503) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0 %505 : Float(1, *, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%503) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %506 : Float(1, *, 46, 60, strides=[24840, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%505) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0 %input.84 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%502, %504, %506, %model.module.backbone.layer3.1.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0 %508 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.84, %model.module.backbone.layer3.1.bn2.weight, %model.module.backbone.layer3.1.bn2.bias, %model.module.backbone.layer3.1.bn2.running_mean, %model.module.backbone.layer3.1.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0 %509 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%508) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %730 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%509, %731, %732) %512 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Add(%730, %499) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0 %513 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%512) %733 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%513, %734, %735) %516 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%733) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0 %517 : Float(1, 27, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%516,

7

This is the 3rd part :

%model.module.backbone.layer3.2.conv2_offset.weight, %model.module.backbone.layer3.2.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %518 : Float(1, 18, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%517) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0   %519 : Float(1, *, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%517) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %520 : Float(1, *, 46, 60, strides=[24840, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%519) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %input.93 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%516, %518, %520, %model.module.backbone.layer3.2.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0   %522 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.93, %model.module.backbone.layer3.2.bn2.weight, %model.module.backbone.layer3.2.bn2.bias, %model.module.backbone.layer3.2.bn2.running_mean, %model.module.backbone.layer3.2.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0   %523 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%522) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %736 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%523, %737, %738)   %526 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Add(%736, %513) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0   %527 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%526)   %739 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%527, %740, %741)   %530 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%739) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %531 : Float(1, 27, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%530, %model.module.backbone.layer3.3.conv2_offset.weight, %model.module.backbone.layer3.3.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %532 : Float(1, 18, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%531) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0   %533 : Float(1, *, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%531) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %534 : Float(1, *, 46, 60, strides=[24840, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%533) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %input.102 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%530, %532, %534, %model.module.backbone.layer3.3.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0   %536 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.102, %model.module.backbone.layer3.3.bn2.weight, %model.module.backbone.layer3.3.bn2.bias, %model.module.backbone.layer3.3.bn2.running_mean, %model.module.backbone.layer3.3.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0   %537 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%536) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %742 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%537, %743, %744)   %540 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Add(%742, %527) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0   %541 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%540)   %745 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%541, %746, %747)   %544 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%745) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %545 : Float(1, 27, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%544, %model.module.backbone.layer3.4.conv2_offset.weight, %model.module.backbone.layer3.4.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %546 : Float(1, 18, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%545) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0   %547 : Float(1, *, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%545) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %548 : Float(1, *, 46, 60, strides=[24840, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%547) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %input.111 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%544, %546, %548, %model.module.backbone.layer3.4.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0   %550 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.111, %model.module.backbone.layer3.4.bn2.weight, %model.module.backbone.layer3.4.bn2.bias, %model.module.backbone.layer3.4.bn2.running_mean, %model.module.backbone.layer3.4.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0   %551 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%550) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %748 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%551, %749, %750)   %554 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Add(%748, %541) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0   %555 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%554)   %751 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%555, %752, %753)   %558 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%751) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %559 : Float(1, 27, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%558, %model.module.backbone.layer3.5.conv2_offset.weight, %model.module.backbone.layer3.5.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %560 : Float(1, 18, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%559) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0   %561 : Float(1, *, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%559) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %562 : Float(1, *, 46, 60, strides=[24840, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%561) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %input.120 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%558, %560, %562, %model.module.backbone.layer3.5.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0   %564 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.120, %model.module.backbone.layer3.5.bn2.weight, %model.module.backbone.layer3.5.bn2.bias, %model.module.backbone.layer3.5.bn2.running_mean, %model.module.backbone.layer3.5.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0   %565 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%564) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %754 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%565, %755, %756)   %568 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Add(%754, %555) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0   %569 : Float(1, 1024, 46, 60, strides=[2826240, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%568) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %757 : Float(1, 512, 46, 60, strides=[1413120, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%569, %758, %759)   %572 : Float(1, 512, 46, 60, strides=[1413120, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%757) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %573 : Float(1, 27, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%572, %model.module.backbone.layer4.0.conv2_offset.weight, %model.module.backbone.layer4.0.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %574 : Float(1, 18, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%573) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0   %575 : Float(1, *, 46, 60, strides=[74520, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%573) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %576 : Float(1, *, 46, 60, strides=[24840, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%575) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %input.129 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 2, 1, 1, 1, 1)(%572, %574, %576, %model.module.backbone.layer4.0.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0   %578 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.129, %model.module.backbone.layer4.0.bn2.weight, %model.module.backbone.layer4.0.bn2.bias, %model.module.backbone.layer4.0.bn2.running_mean, %model.module.backbone.layer4.0.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0   %579 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%578) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %760 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%579, %761, %762)   %763 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[2, 2]](%569, %764, %765)   %584 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Add(%760, %763) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0   %585 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%584)   %766 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%585, %767, %768)   %588 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%766) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %589 : Float(1, 27, 23, 30, strides=[18630, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%588, %model.module.backbone.layer4.1.conv2_offset.weight, %model.module.backbone.layer4.1.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %590 : Float(1, 18, 23, 30, strides=[18630, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%589) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0   %591 : Float(1, *, 23, 30, strides=[18630, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%589) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %592 : Float(1, *, 23, 30, strides=[6210, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%591) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %input.139 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%588, %590, %592, %model.module.backbone.layer4.1.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0   %594 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.139, %model.module.backbone.layer4.1.bn2.weight, %model.module.backbone.layer4.1.bn2.bias, %model.module.backbone.layer4.1.bn2.running_mean, %model.module.backbone.layer4.1.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0   %595 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%594) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %769 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%595, %770, %771)   %598 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Add(%769, %585) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0   %599 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%598)   %772 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%599, %773, %774)   %602 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%772) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %603 : Float(1, 27, 23, 30, strides=[18630, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%602, %model.module.backbone.layer4.2.conv2_offset.weight, %model.module.backbone.layer4.2.conv2_offset.bias) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %604 : Float(1, 18, 23, 30, strides=[18630, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[18], starts=[0]](%603) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:157:0   %605 : Float(1, *, 23, 30, strides=[18630, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Slice[axes=[1], ends=[9223372036854775807], starts=[-9]](%603) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %606 : Float(1, *, 23, 30, strides=[6210, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%605) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:158:0   %input.148 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = ^ModulatedDeformConvFunction(None, 1, 1, 1, 1, 1)(%602, %604, %606, %model.module.backbone.layer4.2.conv2.weight) # D:\Development\work now\emgu_newFeatures\DB\assets\ops\dcn\modules\deform_conv.py:128:0   %608 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%input.148, %model.module.backbone.layer4.2.bn2.weight, %model.module.backbone.layer4.2.bn2.bias, %model.module.backbone.layer4.2.bn2.running_mean, %model.module.backbone.layer4.2.bn2.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0   %609 : Float(1, 512, 23, 30, strides=[353280, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%608) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %775 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%609, %776, %777)   %612 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Add(%775, %599) # D:\Development\work now\emgu_newFeatures\DB\backbones\resnet.py:172:0   %613 : Float(1, 2048, 23, 30, strides=[1413120, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%612) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %614 : Float(1, 256, 23, 30, strides=[176640, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%613, %model.module.decoder.in5.weight) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:396:0   %615 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%569, %model.module.decoder.in4.weight) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:396:0   %616 : Float(1, 256, 92, 120, strides=[2826240, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%483, %model.module.decoder.in3.weight) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:396:0   %617 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%425, %model.module.decoder.in2.weight) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:396:0   %621 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Upsample[mode="nearest"](%614, %781) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:3532:0   %622 : Float(1, 256, 46, 60, strides=[706560, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Add(%621, %615) # D:\Development\work now\emgu_newFeatures\DB\decoders\seg_detector.py:124:0   %626 : Float(1, 256, 92, 120, strides=[2826240, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Upsample[mode="nearest"](%622, %782) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:3532:0   %627 : Float(1, 256, 92, 120, strides=[2826240, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Add(%626, %616) # D:\Development\work now\emgu_newFeatures\DB\decoders\seg_detector.py:125:0   %631 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Upsample[mode="nearest"](%627, %783) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:3532:0   %632 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Add(%631, %617) # D:\Development\work now\emgu_newFeatures\DB\decoders\seg_detector.py:126:0   %633 : Float(1, 64, 23, 30, strides=[44160, 690, 30, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%614, %model.module.decoder.out5.0.weight) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:396:0   %637 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Upsample[mode="nearest"](%633, %784) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:3532:0   %638 : Float(1, 64, 46, 60, strides=[176640, 2760, 60, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%622, %model.module.decoder.out4.0.weight) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:396:0   %642 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Upsample[mode="nearest"](%638, %785) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:3532:0   %643 : Float(1, 64, 92, 120, strides=[706560, 11040, 120, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%627, %model.module.decoder.out3.0.weight) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:396:0   %647 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Upsample[mode="nearest"](%643, %786) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:3532:0   %648 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%632, %model.module.decoder.out2.weight) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:396:0   %649 : Float(1, 256, 184, 240, strides=[11304960, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Concat[axis=1](%637, %642, %647, %648) # D:\Development\work now\emgu_newFeatures\DB\decoders\seg_detector.py:133:0   %778 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%649, %779, %780)   %652 : Float(1, 64, 184, 240, strides=[2826240, 44160, 240, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%778) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %653 : Float(1, 64, 368, 480, strides=[11304960, 176640, 480, 1], requires_grad=1, device=cuda:0) = onnx::ConvTranspose[dilations=[1, 1], group=1, kernel_shape=[2, 2], pads=[0, 0, 0, 0], strides=[2, 2]](%652, %model.module.decoder.binarize.3.weight, %model.module.decoder.binarize.3.bias) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:842:0   %654 : Float(1, 64, 368, 480, strides=[11304960, 176640, 480, 1], requires_grad=1, device=cuda:0) = onnx::BatchNormalization[epsilon=1.0000000000000001e-05, momentum=0.90000000000000002](%653, %model.module.decoder.binarize.4.weight, %model.module.decoder.binarize.4.bias, %model.module.decoder.binarize.4.running_mean, %model.module.decoder.binarize.4.running_var) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:2147:0   %655 : Float(1, 64, 368, 480, strides=[11304960, 176640, 480, 1], requires_grad=1, device=cuda:0) = onnx::Relu(%654) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\functional.py:1204:0   %656 : Float(1, 1, 736, 960, strides=[706560, 706560, 960, 1], requires_grad=1, device=cuda:0) = onnx::ConvTranspose[dilations=[1, 1], group=1, kernel_shape=[2, 2], pads=[0, 0, 0, 0], strides=[2, 2]](%655, %model.module.decoder.binarize.6.weight, %model.module.decoder.binarize.6.bias) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\conv.py:842:0   %output : Float(1, 1, 736, 960, strides=[706560, 706560, 960, 1], requires_grad=1, device=cuda:0) = onnx::Sigmoid(%656) # C:\Program Files (x86)\Microsoft Visual Studio\Shared\Python37_64\lib\site-packages\torch\nn\modules\activation.py:299:0   return (%output)
8

I would strongly suggest formatting that stuff properly. it is not one single long line.

9

Hi Berak ,

I have been changing torch version to try different possibilities but there is always an error . My question now : Do you know if onnx opset version could be changed despite I use a constant pytorch version ? Another way : The only possibility to change the onnex opset version is by changing the pytorch version?

10

that’s what i’d recommend. please try 9 or 11

no, it’s just a parameter with the export() function
but obviously, later pytorch versions support larger opset numbers, no ?

11

I just used this : torch.onnx.export(model, dummy_input, ‘onnx_model_name.onnx’, opset_version=9) and it shows the same : "ONNX export failed: Couldn’t export Python operator Scatter ". The opset_version=9 and It should be compatible, right ? I tried opset_version=11 and the same .

1 Like
12

Hi Berak ,

I decided to comment all the parallelize() method calls in model.py . I remembered someone talking about that so all the parallelism approach will be avoid and the execution will run sequentially. Right now , the Scatter issue disappeared BUT now it shows something similar with : ModulatedDeformConvFunction . Message: ONNX export failed: Couldn’t export Python operator ModulatedDeformConvFunction . Do you have any idea ? I would like to know an exact configuration that had generated the .onnx models that comes from GitHub - MhLiao/DB: A PyTorch implementation of "Real-time Scene Text Detection with Differentiable Binarization". . What Python3 version , Pytorch version, CUDA version does it use?. That would help a lot. If there exist a configuration used with Windows 10 it would be better .
Thanks a lot .

13

seriously, this is all about pytorch, not opencv.

maybe try to ask in a more specific place ?

hrrmm, try to ask the author/owner of that repo ?

14

Yes , I did it . I asked for help to the repo owner the same time I started asking you. I have not received an answer yet. I was trying this repo because Opencv references it. I thought that you had to do the same process that I am doing or at least something similar. Sorry , I apologize. Thanks for your time .

1 Like
15

dont feel sorry, you’ve done nothing wrong at all here.