Openpcdet自定义数据集转onnx

启动docker环境

nvidia-docker run --rm -ti -v /home/$USER/:/home/$USER/ --net=host --rm pointpillar:latest

安装Openpcdet

git clone https://github.com/open-mmlab/OpenPCDet.git
cd OpenPCDet && git checkout 846cf3e && python3 setup.py develop

安装成功

安装CUDA-PointPillars

git clone https://github.com/NVIDIA-AI-IOT/CUDA-PointPillars.git
cd CUDA-PointPillars

转onnx模型

python3 tool/export_onnx.py --ckpt ckpts/checkpoint_epoch_300.pth --out_dir model

报错：

onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input '195' of node:
name: ScatterND_55 OpType: ScatterND
 is not output of any previous nodes.

onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input '195' of node:
name: ScatterND_55 OpType: ScatterND
 is not output of any previous nodes.

解决：

这里降低一下onnx-simplifier的版本

pip install onnx-simplifier==0.4.33

 继续运行后发现又报错：

onnx.onnx_cpp2py_export.checker.ValidationError: Your model ir_version 10 is higher than the checker's (9).

onnx.onnx_cpp2py_export.checker.ValidationError: Your model ir_version 10 is higher than the checker's (9).

 

解决：

降低onnx版本

pip install onnx==1.14.1

再次运行

成功