mmdeployv0.6 mmdetectionv2.4、mmcv-full1.6安装及环境搭建

一、安装条件

Step 1.创建conda虚拟环境并激活

conda create -n mmdeploy python=3.8 -y   #创建环境
conda activate mmdeploy                 #激活环境

Step 2.安装 PyTorch

#如果网不好，可以这样安装
pip3 install torch==1.8.1+cu111 torchvision==0.9.1+cu111   -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
 
#验证是否安装成功
 
>>> import torchvision
>>> import torch
>>> import.__version__
  File "<stdin>", line 1
    import.__version__
          ^
SyntaxError: invalid syntax
>>> torch.__version__
'1.11.0+cu113'

在这里插入图片描述

二、安装mmcv-full 1.6.0

1、pip安装

pip install mmcv-full==1.6.0

2、源码安装

https://github.com/open-mmlab/mmcv.git
cd mmcv
MMCV_WITH_OPS=1 python -m pip install -e . -v

3、验证安装

python .dev_scripts/check_installation.py

在这里插入图片描述

三、安装推理引擎

1、安装ONNXRuntime

安装ONNXRuntime的python包

pip install onnxruntime==1.8.1

2、安装预编译包要求的推理后端 TensorRT

在本例中，我们需要安装 TensorRT（含 cuDNN）推理引擎。因在 NVIDIA 官网下载软件包，必须要登录认证，所以请预先登录并下载所需的 TensorRT 和 cuDNN。请注意： TensorRT 版本、cuDNN 版本要和 CUDA 版本匹配。适用于 x86_64 架构的 TensorRT 8.2 GA Update 2

下载完毕后，您可以参考如下方法安装。这里，我们以 TensorRT 8.2.3.0、cuDNN 8.2 为例：
在这里插入图片描述

# !!! 从 NVIDIA 官网下载 与 cuda toolkit 匹配的 tensorrt 到当前的工作目录
tar -zxvf TensorRT-8.2.3.0.Linux.x86_64-gnu.cuda-11.4.cudnn8.2.tar.gz
cd TensorRT-8.2.3.0
python -m pip install TensorRT-8.2.3.0/python/tensorrt-8.2.3.0-cp37-none-linux_x86_64.whl 
pip install pycuda

# vim ~/.bashrc 配置环境变量，添加下面两行，根据自己的目录修改，最后source ~/.bashrc  使生效。
export TENSORRT_DIR=/home/xbsj/gaoying/lib/TensorRT-8.2.3.0
export LD_LIBRARY_PATH=${TENSORRT_DIR}/lib:$LD_LIBRARY_PATH

二、源码安装mmdeploy v0.6 环境

1、安装 MMDeploy 和 SDK

从 v0.5.0 之后，MMDeploy 开始提供预编译包MMDeploy Release V0.6.0。
在这里插入图片描述

1、MMDeploy-onnxruntime 预编译包

在这里插入图片描述

python -m pip install dist/mmdeploy-0.6.0-py38-none-linux_x86_64.whl
python -m pip install sdk/python/mmdeploy_python-0.6.0-cp38-none-linux_x86_64.whl

2、MMDeploy-TensoRT 预编译包

CuDNN下载地址

您可以根据目标软硬件平台，从这里选择并下载预编译包。在 NVIDIA 设备上，我们推荐使用 MMDeploy-TensoRT 预编译包：

wget https://github.com/open-mmlab/mmdeploy/releases/download/v0.6.0/mmdeploy-0.6.0-linux-x86_64-cuda11.1-tensorrt8.2.3.0.tar.gz

tar -zxvf mmdeploy-0.6.0-linux-x86_64-cuda11.1-tensorrt8.2.3.0.tar.gz

cd mmdeploy-0.6.0-linux-x86_64-cuda11.1-tensorrt8.2.3.0

在这里插入图片描述

python -m pip install dist/mmdeploy-0.6.0-py38-none-linux_x86_64.whl
python -m pip install sdk/python/mmdeploy_python-0.6.0-cp38-none-linux_x86_64.whl
pip install pycuda

# vim ~/.bashrc 配置环境变量，添加下面，根据自己的目录修改，最后source ~/.bashrc  使生效。
export LD_LIBRARY_PATH=/work1/sre/hyliu41/MMproject/mmdeploy-0.6.0-linux-x86_64-cuda11.1-tensorrt8.2.3.0/sdk/lib:$LD_LIBRARY_PATH            # 改为自己的目录
export TENSORRT_DIR=$(pwd)/TensorRT-8.2.3.0
export LD_LIBRARY_PATH=${TENSORRT_DIR}/lib:$LD_LIBRARY_PATH
# !!! Moreover, download cuDNN 8.2.1 CUDA 11.x tar package from NVIDIA, and extract it to the current directory
export CUDNN_DIR=$(pwd)/cuda
export LD_LIBRARY_PATH=$CUDNN_DIR/lib64:$LD_LIBRARY_PATH
cd ..

在这里插入图片描述

3、安装mmdetectionv2.4

在这里插入图片描述

https://github.com/open-mmlab/mmdetection.git
cd mmdetection
pip install -r requirements/build.txt
pip install -v -e .

三、模型转换与推理

1、Faster RCNN

在准备工作就绪后，我们可以使用 MMDeploy 中的工具 deploy.py，将 OpenMMLab 的 PyTorch 模型转换成推理后端支持的格式。

以 MMDetection 中的 Faster R-CNN 为例，我们可以使用如下命令，将 PyTorch 模型转换成可部署在 NVIDIA GPU 上的 TenorRT 模型：

# 克隆 mmdeploy 仓库。转换时，需要使用 mmdeploy 仓库中的配置文件，建立转换流水线
git clone --recursive https://github.com/open-mmlab/mmdeploy.git
python -m pip install -r mmdeploy/requirements/runtime.txt


# 克隆 mmdetection 仓库。转换时，需要使用 mmdetection 仓库中的模型配置文件，构建 PyTorch nn module
python -m pip install mmdet==2.24.0
git clone https://github.com/open-mmlab/mmdetection.git

# 下载 Faster R-CNN 模型权重
cd ./mmdetection/checkpoints
wget https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth -O faster_rcnn_r50_fpn_1x_coco.pth

# 设置工作路径
cd ../../mmdeploy
mkdir mmdeploy_models
mkdir mmdeploy_models/faster_rcnn

# 执行转换命令，实现端到端的转换[[模板]]
python ${MMDEPLOY_DIR}/tools/deploy.py \
    ${MMDEPLOY_DIR}/configs/mmdet/detection/detection_tensorrt_dynamic-320x320-1344x1344.py \
    ${MMDET_DIR}/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py \
    ${CHECKPOINT_DIR}/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
    ${MMDET_DIR}/demo/demo.jpg \
    --work-dir ${WORK_DIR} \
    --device cuda:0 \
    --dump-info

# 执行转换命令，实现端到端的转换
python tools/deploy.py configs/mmdet/detection/detection_tensorrt_dynamic-320x320-1344x1344.py ../mmdetection/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py ../mmdetection/checkpoints/faster_rcnn_r50_fpn_1x_coco.pth ../mmdetection/demo/demo.jpg --work-dir mmdeploy_models/faster_rcnn/ --device cuda:0 --dump-info

您可以将上述模型转换为 onnx 模型并执行 ONNX 运行时推理，只需将“detection_tensorrt_dynamic-320x320-1344x1344.py”更改为“detection_onnxruntime_dynamic.py”并将“–device”设为“cpu”即可。

1、问题

在这里插入图片描述

2、推理模型

模型转换后，我们不仅可以通过模型转换器进行推理，还可以通过推理SDK进行推理。
（1）通过模型转换器进行推理
模型转换器提供了统一的inference_model API 来完成这项工作，使所有推理后端 API 对用户透明。以之前转换的 Faster R-CNN tensorrt 模型为例

2、yolov3

在中找到模型的代码库文件夹configs/。要转换 yolov3 模型，您需要检查configs/mmdet文件夹。

在中找到模型的任务文件夹configs/codebase_folder/。对于 yolov3 模型，您需要检查configs/mmdet/detection文件夹。

在中找到部署配置文件configs/codebase_folder/task_folder/。要将 yolov3 模型部署到 onnx 后端，您可以使用configs/mmdet/detection/detection_onnxruntime_dynamic.py。

python ./tools/deploy.py \
    configs/mmdet/detection/detection_tensorrt_dynamic-320x320-1344x1344.py \
    $PATH_TO_MMDET/configs/yolo/yolov3_d53_8xb8-ms-608-273e_coco.py \
    $PATH_TO_MMDET/checkpoints/yolo/yolov3_d53_mstrain-608_273e_coco_20210518_115020-a2c3acb8.pth \
    $PATH_TO_MMDET/demo/demo.jpg \
    --work-dir work_dir \
    --show \
    --device cuda:0

${MMDEPLOY_DIR}/tools/deploy.py 是一个方便模型转换的工具。您可以阅读如何转换模型了解更多细节。

detection_tensorrt_dynamic-320x320-1344x1344.py 是一个参数配置文件。该文件的命名遵循如下规则：

<任务名>_<推理后端>-[后端特性]_<动态模型支持>.py

python tools/check_env.py

在这里插入图片描述

四、参考博客

【mmdeploy】【TODO】使用mmdeploy将mmdetection模型转tensorrt
6.6.tensorRT高级(1)-mmdetection框架下yolox模型导出并推理
mmdetection模型转onnx和tensorrt实战
MMDeploy部署实战系列
windows环境下使用mmdetection+mmdeploy训练自定义数据集并转成onnx格式部署
【环境搭建：onnx模型部署】onnxruntime-gpu安装与测试（python）
mmdeploy windows下pth2onnx 并对onnx进行量化

四、问题

添加链接描述

在这里插入图片描述

问题1、由于未注册，无法对 ONNX 模型运行评估MMCVDeformConv2d

转换为 ONNX 的日志：

(llens) ➜  experiments sh execute.sh
[2022-05-05 21:18:00.119] [mmdeploy] [info] [model.cpp:95] Register 'DirectoryModel'
/workspace/model/mmdetection/mmdet/datasets/utils.py:70: UserWarning: "ImageToTensor" pipeline is replaced by "DefaultFormatBundle" for batch inference. It is recommended to manually replace it in the test data pipeline in your config file.
  'data pipeline in your config file.', UserWarning)
2022-05-05 21:18:01,365 - mmdeploy - INFO - torch2onnx start.
[2022-05-05 21:18:03.182] [mmdeploy] [info] [model.cpp:95] Register 'DirectoryModel'
load checkpoint from local path: /workspace/experiments/deploy_test_0505/best_instance_mAP_epoch_18.pth
/workspace/model/mmdetection/mmdet/datasets/utils.py:70: UserWarning: "ImageToTensor" pipeline is replaced by "DefaultFormatBundle" for batch inference. It is recommended to manually replace it in the test data pipeline in your config file.
  'data pipeline in your config file.', UserWarning)
2022-05-05 21:18:05,391 - mmdeploy - WARNING - DeprecationWarning: get_onnx_config will be deprecated in the future. 
2022-05-05:21:18:05,mmdeploy WARNING  [utils.py:92] DeprecationWarning: get_onnx_config will be deprecated in the future. 
/workspace/mmdeploy/mmdeploy/core/optimizers/function_marker.py:158: TracerWarning: Converting a tensor to a Python integer might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  ys_shape = tuple(int(s) for s in ys.shape)
/opt/conda/envs/llens/lib/python3.7/site-packages/torch/nn/functional.py:2359: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
  _verify_batch_size([input.size(0) * input.size(1) // num_groups, num_groups] + list(input.size()[2:]))
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/deform_conv.py:299: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  input_pad = (x.size(2) < self.kernel_size[0]) or (x.size(3) <
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/deform_conv.py:301: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if input_pad:
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/deform_conv.py:181: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
  output_size += ((in_size + (2 * pad) - kernel) // stride_ + 1, )
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/deform_conv.py:182: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if not all(map(lambda s: s > 0, output_size)):
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/deform_conv.py:89: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  cur_im2col_step = min(ctx.im2col_step, input.size(0))
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/deform_conv.py:91: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  ) == 0, 'batch size must be divisible by im2col_step'
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/deform_conv.py:310: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if input_pad:
/opt/conda/envs/llens/lib/python3.7/site-packages/torch/functional.py:445: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at  ../aten/src/ATen/native/TensorShape.cpp:2157.)
  return _VF.meshgrid(tensors, **kwargs)  # type: ignore[attr-defined]
/workspace/mmdeploy/mmdeploy/codebase/mmdet/models/dense_heads/base_dense_head.py:95: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert cls_score.size()[-2:] == bbox_pred.size()[-2:]
/workspace/mmdeploy/mmdeploy/pytorch/functions/topk.py:28: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
  k = torch.tensor(k, device=input.device, dtype=torch.long)
/workspace/mmdeploy/mmdeploy/pytorch/functions/topk.py:33: TracerWarning: Converting a tensor to a Python integer might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  return ctx.origin_func(input, k, dim=dim, largest=largest, sorted=sorted)
/workspace/model/mmdetection/mmdet/core/bbox/coder/distance_point_bbox_coder.py:58: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert points.size(0) == pred_bboxes.size(0)
/workspace/model/mmdetection/mmdet/core/bbox/coder/distance_point_bbox_coder.py:59: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert points.size(-1) == 2
/workspace/model/mmdetection/mmdet/core/bbox/coder/distance_point_bbox_coder.py:60: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert pred_bboxes.size(-1) == 4
/workspace/mmdeploy/mmdeploy/codebase/mmdet/core/post_processing/bbox_nms.py:92: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
  iou_threshold = torch.tensor([iou_threshold], dtype=torch.float32)
/workspace/mmdeploy/mmdeploy/codebase/mmdet/core/post_processing/bbox_nms.py:93: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
  score_threshold = torch.tensor([score_threshold], dtype=torch.float32)
/workspace/mmdeploy/mmdeploy/mmcv/ops/nms.py:40: TracerWarning: Converting a tensor to a Python float might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  score_threshold = float(score_threshold)
/workspace/mmdeploy/mmdeploy/mmcv/ops/nms.py:41: TracerWarning: Converting a tensor to a Python float might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  iou_threshold = float(iou_threshold)
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/nms.py:160: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert boxes.size(1) == 4
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/nms.py:161: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert boxes.size(0) == scores.size(0)
/opt/conda/envs/llens/lib/python3.7/site-packages/mmcv/ops/nms.py:29: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if max_num > 0:
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
/opt/conda/envs/llens/lib/python3.7/site-packages/torch/onnx/symbolic_opset9.py:2819: UserWarning: Exporting aten::index operator of advanced indexing in opset 11 is achieved by combination of multiple ONNX operators, including Reshape, Transpose, Concat, and Gather. If indices include negative values, the exported graph will produce incorrect results.
  "If indices include negative values, the exported graph will produce incorrect results.")
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
WARNING: The shape inference of mmdeploy::MMCVDeformConv2d type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function.
2022-05-05 21:18:29,953 - mmdeploy - INFO - torch2onnx success.
2022-05-05 21:18:29,954 - mmdeploy - WARNING - "visualize_model" has been skipped may be because it's             running on a headless device.
2022-05-05 21:18:29,954 - mmdeploy - INFO - All process success.