how to correctly inference tensorRT model?

[sylivahf]

here, i have the correct results with pytorch model. but get empty with tensorRT model. so, how to modify？
======== torch模型推理 ============
[{'boxes_3d': LiDARInstance3DBoxes(
tensor([[23.8085, -9.0607, -1.9104, 0.7693, 0.8013, 1.7510, 4.1653],
[39.1756, -9.7332, -1.5768, 0.6849, 0.7169, 1.6472, 0.1994],
[20.7223, -8.7755, -1.6595, 3.7742, 1.5835, 1.7513, -0.3699],
[14.9589, -1.0985, -1.3632, 4.2891, 1.8484, 1.6975, -0.2781],
[33.7720, -7.1122, -1.2845, 4.5438, 1.8019, 1.6231, -0.2848],
[ 8.4082, 1.2065, -1.7101, 4.4612, 1.8056, 1.8814, 2.9314],
[ 7.1448, -4.0764, -1.8516, 4.5572, 1.7226, 1.5386, -0.2915],
[ 3.5187, 2.9281, -1.7771, 4.5677, 1.9608, 1.7604, 2.8909]])), 'scores_3d': tensor([0.2288, 0.2085, 0.8782, 0.6951, 0.5204, 0.4221, 0.3904, 0.3322]), 'labels_3d': tensor([0, 0, 2, 2, 2, 2, 2, 2])}]
======== tensorRT模型推理 ============
[[{'boxes_3d': LiDARInstance3DBoxes(
tensor([], size=(0, 7))), 'scores_3d': tensor([]), 'labels_3d': tensor([], dtype=torch.int64)}]]

• about envs：
  2022-10-19 03:37:05,970 - mmdeploy - INFO -

2022-10-19 03:37:05,970 - mmdeploy - INFO - Environmental information
fatal: detected dubious ownership in repository at '/mmdeploy'
To add an exception for this directory, call:

    git config --global --add safe.directory /mmdeploy

2022-10-19 03:37:06,224 - mmdeploy - INFO - sys.platform: linux
2022-10-19 03:37:06,224 - mmdeploy - INFO - Python: 3.8.13 (default, Mar 28 2022, 11:38:47) [GCC 7.5.0]
2022-10-19 03:37:06,224 - mmdeploy - INFO - CUDA available: True
2022-10-19 03:37:06,224 - mmdeploy - INFO - GPU 0: NVIDIA GeForce RTX 3060
2022-10-19 03:37:06,224 - mmdeploy - INFO - CUDA_HOME: /usr/local/cuda
2022-10-19 03:37:06,224 - mmdeploy - INFO - NVCC: Cuda compilation tools, release 11.6, V11.6.124
2022-10-19 03:37:06,224 - mmdeploy - INFO - GCC: gcc (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
2022-10-19 03:37:06,224 - mmdeploy - INFO - PyTorch: 1.8.0
2022-10-19 03:37:06,224 - mmdeploy - INFO - PyTorch compiling details: PyTorch built with:

• GCC 7.3
• C++ Version: 201402
• Intel(R) oneAPI Math Kernel Library Version 2022.0-Product Build 20211112 for Intel(R) 64 architecture applications
• Intel(R) MKL-DNN v1.7.0 (Git Hash 7aed236906b1f7a05c0917e5257a1af05e9ff683)
• OpenMP 201511 (a.k.a. OpenMP 4.5)
• NNPACK is enabled
• CPU capability usage: AVX2
• CUDA Runtime 11.1
• NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-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_37,code=compute_37
• CuDNN 8.0.5
• Magma 2.5.2
• Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.1, CUDNN_VERSION=8.0.5, CXX_COMPILER=/opt/rh/devtoolset-7/root/usr/bin/c++, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -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=1.8.0, 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,

2022-10-19 03:37:06,224 - mmdeploy - INFO - TorchVision: 0.9.0
2022-10-19 03:37:06,224 - mmdeploy - INFO - OpenCV: 4.6.0
2022-10-19 03:37:06,224 - mmdeploy - INFO - MMCV: 1.6.0
2022-10-19 03:37:06,224 - mmdeploy - INFO - MMCV Compiler: GCC 7.3
2022-10-19 03:37:06,224 - mmdeploy - INFO - MMCV CUDA Compiler: 11.1
2022-10-19 03:37:06,224 - mmdeploy - INFO - MMDeploy: 0.9.0+
2022-10-19 03:37:06,224 - mmdeploy - INFO -

2022-10-19 03:37:06,224 - mmdeploy - INFO - Backend information
2022-10-19 03:37:06,537 - mmdeploy - INFO - onnxruntime: 1.8.1 ops_is_avaliable : True
2022-10-19 03:37:06,555 - mmdeploy - INFO - tensorrt: 8.2.4.2 ops_is_avaliable : True
2022-10-19 03:37:06,576 - mmdeploy - INFO - ncnn: None ops_is_avaliable : False
2022-10-19 03:37:06,577 - mmdeploy - INFO - pplnn_is_avaliable: False
2022-10-19 03:37:06,578 - mmdeploy - INFO - openvino_is_avaliable: False
2022-10-19 03:37:06,591 - mmdeploy - INFO - snpe_is_available: False
2022-10-19 03:37:06,592 - mmdeploy - INFO - ascend_is_available: False
2022-10-19 03:37:06,593 - mmdeploy - INFO - coreml_is_available: False
2022-10-19 03:37:06,593 - mmdeploy - INFO -

2022-10-19 03:37:06,593 - mmdeploy - INFO - Codebase information
2022-10-19 03:37:06,594 - mmdeploy - INFO - mmdet: 2.14.0
2022-10-19 03:37:06,594 - mmdeploy - INFO - mmseg: 0.14.1
2022-10-19 03:37:06,594 - mmdeploy - INFO - mmcls: None
2022-10-19 03:37:06,594 - mmdeploy - INFO - mmocr: None
2022-10-19 03:37:06,594 - mmdeploy - INFO - mmedit: None
2022-10-19 03:37:06,594 - mmdeploy - INFO - mmdet3d: 1.0.0rc4
2022-10-19 03:37:06,594 - mmdeploy - INFO - mmpose: None
2022-10-19 03:37:06,594 - mmdeploy - INFO - mmrotate: None

[tpoisonooo]

For TRT backend, check these issues:

• myelin::ir::tensor_t*& myelin::ir::operand_t::tensor() NVIDIA/TensorRT#1541
• Tensorrt output NaN while onnxruntime is fine NVIDIA/TensorRT#2338

In short, you have to upgrade TRT version to docker 8.5 / cu102+8.4

[sylivahf]

你好，感谢回答。我是不是可以理解为，其实主要是cuda版本导致的。因为我之前有升级过trt版本到8.4，升级cuda11.3，问题依然在。