Error when running export_rtdetr_ultralytics:Unsupported: ONNX export of operator get_pool_ceil_padding, input size not accessible

[tms2003]

this is my command:

python3 export_rtdetr_ultralytics.py -w rtdetr-l.pt --simplify --batch 4 -s 640

and got this error:

python3 export_rtdetr_ultralytics.py -w rtdetr-l.pt  --simplify --batch 4 -s 640

Starting: rtdetr-l.pt
Opening RT-DETR Ultralytics model

Ultralytics YOLOv8.2.73 🚀 Python-3.10.12 torch-2.1.1+cu121 CPU (Intel Core(TM) i5-4590 3.30GHz)
rt-detr-l summary: 494 layers, 32,148,140 parameters, 0 gradients, 103.8 GFLOPs

Creating labels.txt file

Exporting the model to ONNX
Traceback (most recent call last):
  File "/opt/nvidia/deepstream/ultralytics/export_rtdetr_ultralytics.py", line 124, in <module>
    sys.exit(main(args))
  File "/opt/nvidia/deepstream/ultralytics/export_rtdetr_ultralytics.py", line 92, in main
    torch.onnx.export(model, onnx_input_im, onnx_output_file, verbose=False, opset_version=args.opset,
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py", line 516, in export
    _export(
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py", line 1596, in _export
    graph, params_dict, torch_out = _model_to_graph(
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py", line 1139, in _model_to_graph
    graph = _optimize_graph(
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py", line 677, in _optimize_graph
    graph = _C._jit_pass_onnx(graph, operator_export_type)
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py", line 1940, in _run_symbolic_function
    return symbolic_fn(graph_context, *inputs, **attrs)
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/symbolic_helper.py", line 395, in wrapper
    return fn(g, *args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/symbolic_helper.py", line 306, in wrapper
    return fn(g, *args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/symbolic_opset10.py", line 203, in symbolic_fn
    padding_ceil = opset9.get_pool_ceil_padding(
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/symbolic_opset9.py", line 1565, in get_pool_ceil_padding
    return symbolic_helper._unimplemented(
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/symbolic_helper.py", line 612, in _unimplemented
    _onnx_unsupported(f"{op}, {msg}", value)
  File "/usr/local/lib/python3.10/dist-packages/torch/onnx/symbolic_helper.py", line 623, in _onnx_unsupported
    raise errors.SymbolicValueError(
torch.onnx.errors.SymbolicValueError: Unsupported: ONNX export of operator get_pool_ceil_padding, input size not accessible. Please feel free to request support or submit a pull request on PyTorch GitHub: https://github.com/pytorch/pytorch/issues  [Caused by the value 'input.4 defined in (%input.4 : Float(*, *, *, *, strides=[3297312, 103041, 321, 1], requires_grad=0, device=cpu) = onnx::Pad[mode="constant"](%449, %476, %452), scope: torch.nn.modules.container.Sequential::/ultralytics.nn.tasks.DetectionModel::0/ultralytics.nn.modules.block.HGStem::model.0 # /opt/nvidia/deepstream/ultralytics/ultralytics/nn/modules/block.py:114:0
)' (type 'Tensor') in the TorchScript graph. The containing node has kind 'onnx::Pad'.] 
    (node defined in /opt/nvidia/deepstream/ultralytics/ultralytics/nn/modules/block.py(114): forward
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1508): _slow_forward
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1527): _call_impl
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1518): _wrapped_call_impl
/opt/nvidia/deepstream/ultralytics/ultralytics/nn/tasks.py(141): _predict_once
/opt/nvidia/deepstream/ultralytics/ultralytics/nn/tasks.py(120): predict
/opt/nvidia/deepstream/ultralytics/ultralytics/nn/tasks.py(102): forward
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1508): _slow_forward
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1527): _call_impl
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1518): _wrapped_call_impl
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/container.py(215): forward
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1508): _slow_forward
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1527): _call_impl
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1518): _wrapped_call_impl
/usr/local/lib/python3.10/dist-packages/torch/jit/_trace.py(124): wrapper
/usr/local/lib/python3.10/dist-packages/torch/jit/_trace.py(133): forward
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1527): _call_impl
/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py(1518): _wrapped_call_impl
/usr/local/lib/python3.10/dist-packages/torch/jit/_trace.py(1285): _get_trace_graph
/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py(915): _trace_and_get_graph_from_model
/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py(1011): _create_jit_graph
/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py(1135): _model_to_graph
/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py(1596): _export
/usr/local/lib/python3.10/dist-packages/torch/onnx/utils.py(516): export
/opt/nvidia/deepstream/ultralytics/export_rtdetr_ultralytics.py(92): main
/opt/nvidia/deepstream/ultralytics/export_rtdetr_ultralytics.py(124): <module>
)

    Inputs:
        #0: 449 defined in (%449 : Float(4, 32, 320, 320, strides=[3276800, 102400, 320, 1], requires_grad=0, device=cpu) = onnx::Relu(%input), scope: torch.nn.modules.container.Sequential::/ultralytics.nn.tasks.DetectionModel::0/ultralytics.nn.modules.block.HGStem::model.0/ultralytics.nn.modules.conv.Conv::stem1/torch.nn.modules.activation.ReLU::act # /usr/local/lib/python3.10/dist-packages/torch/nn/functional.py:1469:0
    )  (type 'Tensor')
        #1: 476 defined in (%476 : Long(8, strides=[1], device=cpu) = onnx::Cast[to=7](%475), scope: torch.nn.modules.container.Sequential::/ultralytics.nn.tasks.DetectionModel::0/ultralytics.nn.modules.block.HGStem::model.0 # /opt/nvidia/deepstream/ultralytics/ultralytics/nn/modules/block.py:114:0
    )  (type 'Tensor')
        #2: 452 defined in (%452 : NoneType = prim::Constant(), scope: torch.nn.modules.container.Sequential::/ultralytics.nn.tasks.DetectionModel::0/ultralytics.nn.modules.block.HGStem::model.0
    )  (type 'NoneType')
    Outputs:
        #0: input.4 defined in (%input.4 : Float(*, *, *, *, strides=[3297312, 103041, 321, 1], requires_grad=0, device=cpu) = onnx::Pad[mode="constant"](%449, %476, %452), scope: torch.nn.modules.container.Sequential::/ultralytics.nn.tasks.DetectionModel::0/ultralytics.nn.modules.block.HGStem::model.0 # /opt/nvidia/deepstream/ultralytics/ultralytics/nn/modules/block.py:114:0
    )  (type 'Tensor')

My environment is:

onnx                     1.16.2
onnxruntime              1.18.1
onnxsim                  0.4.36

[prashant-dn]

When I try to export with a custom size python3 rtdetr_export.py -w best.pt -s 384 640 --simplify --dynamic It works

[marcoslucianops]

Use different torch version.