I installed the dcn-op-v2 and it can run successfully, but when i want to export the onnx model, it got the error above.

Thank you very much for your materials. I managed to use your pytorch dcn to train my model. However, when I convert this model to ONNX format, this error apprears. Do you know why?

  File "/data1/hoangbm/codes/CenterNet/src/lib/models/networks/dcnv2_torch15/dcn_v2_wrapper.py", line 105, in forward
    return DeformableConv2DFunction.apply(inputs, self.weight, self.bias, offset, mask, self.stride, self.pad, self.dilation, self.deformable_groups)
  File "/data1/hoangbm/codes/CenterNet/src/lib/models/networks/dcnv2_torch15/dcn_v2_wrapper.py", line 28, in forward
    ctx.deformable_groups
IndexError: _Map_base::at

This code produce this error

onnx.export(model, input_, '/data1/hoangbm/temp/pp_cnet15_dla.onnx', 
            keep_initializers_as_inputs=True,
            verbose=False, operator_export_type=OperatorExportTypes.ONNX)

Hi @SsisyphusTao, great job!
I found that if one uses batch_size greater than 1, the only first tensor in batch is correct, the others are zero or nearly zero.

For example, if change several lines of code in examples/mbv3_centernet_trt7.py let's say
at line 216 builder.max_batch_size = 4
at line 225 img = np.random.randn(4,3,224,224).astype(np.float32)
at line 254 [hm, wh, reg, _] = common.do_inference(context, bindings=bindings, inputs=inputs, outputs=outputs, stream=stream, batch_size=4)

And try inference on batch size 4, the ouputs are correct only for first image in the batch.
The other outputs are zeros. I added print(hm[-200:]) to check and
as the output is:
[ 0.0000000e+00 1.7881393e-06 2.3245811e-06 1.4901161e-06 5.9604645e-08 -1.1920929e-07 -1.1920929e-07 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 -2.9802322e-07 -1.1920929e-07 1.7881393e-07 0.0000000e+00 0.0000000e+00 1.9490719e-05 2.5033951e-06 2.3603439e-05 4.7683716e-07 -1.0728836e-06 1.9669533e-06 0.0000000e+00 0.0000000e+00 -1.4901161e-06 5.9604645e-08 -3.5762787e-07 4.1127205e-05 3.9935112e-06 5.0365925e-05 0.0000000e+00 0.0000000e+00 0.0000000e+00 1.1920929e-07 9.5367432e-07 -1.1920929e-06 -2.2649765e-06 -3.3378601e-06 0.0000000e+00 0.0000000e+00 -3.2782555e-06 -2.3841858e-07 -2.5629997e-06 -3.9935112e-06 -7.0929527e-06 4.1723251e-07 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 1.1920929e-07 -2.9802322e-07 0.0000000e+00 0.0000000e+00 7.7486038e-07 -5.3644180e-07 -5.9604645e-08 2.3841858e-07 4.1723251e-07 2.0265579e-06 0.0000000e+00 0.0000000e+00 -2.9802322e-07 -5.9604645e-08 1.7881393e-07 1.7881393e-06 -1.3113022e-06 -5.9604645e-08 0.0000000e+00 0.0000000e+00 0.0000000e+00 -5.9604645e-08 0.0000000e+00 5.3644180e-07 5.9604645e-08 1.7881393e-07 0.0000000e+00 0.0000000e+00 1.3709068e-06 -1.2516975e-06 -7.1525574e-07 9.5367432e-07 -2.3365021e-05 4.9471855e-06 0.0000000e+00 0.0000000e+00 -2.9802322e-07 1.1920929e-07 5.9604645e-08 1.7881393e-07 -2.3841858e-07 2.3841858e-07 0.0000000e+00 0.0000000e+00 -3.7550926e-06 7.1525574e-07 -1.7285347e-06 -2.2053719e-06 1.3113022e-06 3.9339066e-06 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 -4.7683716e-07 -1.4901161e-06 -2.9802322e-07 0.0000000e+00 0.0000000e+00 -2.1457672e-06 -2.3841858e-07 5.9604645e-07 -1.0132790e-06 5.9604645e-08 7.1525574e-07 0.0000000e+00 0.0000000e+00 -1.1920929e-07 4.7683716e-07 -4.1723251e-07 -2.2053719e-06 1.4305115e-06 1.1324883e-06 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 9.2387199e-06 -7.0333481e-06 0.0000000e+00 0.0000000e+00 0.0000000e+00 1.7881393e-07 2.9802322e-07 -4.1723251e-07 -1.0251999e-05 -4.2915344e-06 1.8775463e-05 0.0000000e+00 0.0000000e+00 6.5565109e-07 -2.9802322e-07 6.5565109e-07 0.0000000e+00 -5.9604645e-07 -5.9604645e-07 0.0000000e+00 0.0000000e+00 1.8477440e-06 -2.6226044e-06 -1.7881393e-06 -2.9802322e-07 -1.7881393e-07 6.5565109e-07 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 -4.7683716e-07 2.3841858e-07 6.5565109e-07 0.0000000e+00 0.0000000e+00 -7.1525574e-07 5.9604645e-07 -6.5565109e-07 1.7881393e-07 -1.6093254e-06 1.1324883e-06 0.0000000e+00 0.0000000e+00 1.1920929e-07 -6.5565109e-07 5.9604645e-08 8.9406967e-07 -1.2516975e-06 6.5565109e-07 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 0.0000000e+00 -3.7550926e-06 2.3841858e-07 -3.6954880e-06 0.0000000e+00 0.0000000e+00 0.0000000e+00 -1.1920929e-07 -1.1920929e-07 -1.8775463e-05 -6.1988831e-06 -1.3768673e-05 0.0000000e+00]
May be you know the problem and can point me in right direction? @SsisyphusTao

Hi,

thanks a lot for this great repo. Very helpful!
We would like to use the DCNv2 plugin with tensorrt8 but it appears that is not yet supported yet. We are getting several errors when compiling.

Do you plan on upgrading the DCNv2 plugin to support tensorr8 as well or have any preliminary points/work that could help speed up the progress?

Thanks!