[Bug] Warning: The shape inference of mmdeploy::Mark type is missing

[kminoda]

Checklist

• I have searched related issues but cannot get the expected help.
• 2. I have read the FAQ documentation but cannot get the expected help.
• 3. The bug has not been fixed in the latest version.

Describe the bug

I am trying to get partitioned ONNX file for my custom model based on CenterPoint via partition_cfg. Although I can get partitioned ONNX model via tools/deploy.py, some of the shapes are not properly inserted.

What we expect: (ONNX file from our own script)

What we get from mmdeploy:

Also, I get the following warnings, which seems to be the cause of the above issue.

[W shape_type_inference.cpp:1973] Warning: The shape inference of mmdeploy::Mark type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
 (function UpdateReliable)                                                                                                                                                                                         
[W shape_type_inference.cpp:1973] Warning: The shape inference of mmdeploy::Mark type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
 (function UpdateReliable)                                                                                                                                                                                         
[W shape_type_inference.cpp:1973] Warning: The shape inference of mmdeploy::Mark type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
 (function UpdateReliable)  
...

Reproduction

Run tools/deploy.py with following config:

codebase_config = dict(
    type='mmdet3d', task='VoxelDetection', model_type='end2end')

backend_config = dict(
    type='tensorrt',
    common_config=dict(max_workspace_size=1 << 32),
    model_inputs=[
        ...
    ]
)

partition_config = dict(
    type='centerpoint_without_voxelization', # the partition policy name
    apply_marks=True, # should always be set to True
    partition_cfg=[
        dict(
            save_file='pts_backbone_neck_head_centerpoint.onnx', # filename to save the partitioned onnx model
            start=['pillar_feature_net:input'], # [mark_name:input/output, ...]
            end=['pillar_feature_net:output'],  # [mark_name:input/output, ...]
            output_names=['pillar_features'],
            dynamic_axes={
                "input_features": {
                    0: "num_voxels",
                    1: "num_max_points",
                },
                "pillar_features": {
                    0: "num_voxels",
                }
            },
        ),
    ],
)

onnx_config = dict(
    type='onnx',
    export_params=True,
    keep_initializers_as_inputs=False,
    opset_version=11,
    save_file='end2end.onnx',
    ...
)

Environment

07/17 03:58:54 - mmengine - INFO - 

07/17 03:58:54 - mmengine - INFO - **********Environmental information**********
07/17 03:58:55 - mmengine - INFO - sys.platform: linux
07/17 03:58:55 - mmengine - INFO - Python: 3.10.14 (main, Mar 21 2024, 16:24:04) [GCC 11.2.0]
07/17 03:58:55 - mmengine - INFO - CUDA available: True
07/17 03:58:55 - mmengine - INFO - MUSA available: False
07/17 03:58:55 - mmengine - INFO - numpy_random_seed: 2147483648
07/17 03:58:55 - mmengine - INFO - GPU 0: NVIDIA GeForce RTX 4090
07/17 03:58:55 - mmengine - INFO - CUDA_HOME: /usr/local/cuda
07/17 03:58:55 - mmengine - INFO - NVCC: Cuda compilation tools, release 12.1, V12.1.105
07/17 03:58:55 - mmengine - INFO - GCC: gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
07/17 03:58:55 - mmengine - INFO - PyTorch: 2.2.2
07/17 03:58:55 - mmengine - INFO - PyTorch compiling details: PyTorch built with:
  - GCC 9.3
  - C++ Version: 201703
  - Intel(R) oneAPI Math Kernel Library Version 2023.1-Product Build 20230303 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v3.3.2 (Git Hash 2dc95a2ad0841e29db8b22fbccaf3e5da7992b01)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - CUDA Runtime 12.1
  - NVCC architecture flags: -gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90
  - CuDNN 8.9.2
  - Magma 2.6.1
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=12.1, CUDNN_VERSION=8.9.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -D_GLIBCXX_USE_CXX11_ABI=0 -fabi-version=11 -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=pedantic -Wno-error=old-style-cast -Wno-missing-braces -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=2.2.2, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, 

07/17 03:58:55 - mmengine - INFO - TorchVision: 0.17.2
07/17 03:58:55 - mmengine - INFO - OpenCV: 4.9.0
07/17 03:58:55 - mmengine - INFO - MMEngine: 0.10.3
07/17 03:58:55 - mmengine - INFO - MMCV: 2.1.0
07/17 03:58:55 - mmengine - INFO - MMCV Compiler: GCC 11.4
07/17 03:58:55 - mmengine - INFO - MMCV CUDA Compiler: 12.1
07/17 03:58:55 - mmengine - INFO - MMDeploy: 1.3.1+
07/17 03:58:55 - mmengine - INFO - 

07/17 03:58:55 - mmengine - INFO - **********Backend information**********
07/17 03:58:55 - mmengine - INFO - tensorrt:	None
07/17 03:58:55 - mmengine - INFO - ONNXRuntime:	None
07/17 03:58:55 - mmengine - INFO - pplnn:	None
07/17 03:58:55 - mmengine - INFO - ncnn:	None
07/17 03:58:55 - mmengine - INFO - snpe:	None
07/17 03:58:55 - mmengine - INFO - openvino:	None
07/17 03:58:55 - mmengine - INFO - torchscript:	2.2.2
07/17 03:58:55 - mmengine - INFO - torchscript custom ops:	NotAvailable
07/17 03:58:55 - mmengine - INFO - rknn-toolkit:	None
07/17 03:58:55 - mmengine - INFO - rknn-toolkit2:	None
07/17 03:58:55 - mmengine - INFO - ascend:	None
07/17 03:58:55 - mmengine - INFO - coreml:	None
07/17 03:58:55 - mmengine - INFO - tvm:	None
07/17 03:58:55 - mmengine - INFO - vacc:	None
07/17 03:58:55 - mmengine - INFO - 

07/17 03:58:55 - mmengine - INFO - **********Codebase information**********
07/17 03:58:55 - mmengine - INFO - mmdet:	3.2.0
07/17 03:58:55 - mmengine - INFO - mmseg:	None
07/17 03:58:55 - mmengine - INFO - mmpretrain:	None
07/17 03:58:55 - mmengine - INFO - mmocr:	None
07/17 03:58:55 - mmengine - INFO - mmagic:	None
07/17 03:58:55 - mmengine - INFO - mmdet3d:	1.4.0
07/17 03:58:55 - mmengine - INFO - mmpose:	None
07/17 03:58:55 - mmengine - INFO - mmrotate:	None
07/17 03:58:55 - mmengine - INFO - mmaction:	None
07/17 03:58:55 - mmengine - INFO - mmrazor:	None
07/17 03:58:55 - mmengine - INFO - mmyolo:	None

Error traceback

No response

[kminoda]

I am also seeing this warning:

07/17 04:09:42 - mmengine - WARNING - Failed to search registry with scope "mmdet3d" in the "Codebases" registry tree. As a workaround, the current "Codebases" registry in "mmdeploy" is used to build instance. T
his may cause unexpected failure when running the built modules. Please check whether "mmdet3d" is a correct scope, or whether the registry is initialized.