mmdeploy
mmdeploy copied to clipboard
open-mmlab
ONNX parser isn't working with official TensorRT onnx parser
Checklist
- I have searched related issues but cannot get the expected help.
- I have read the FAQ documentation but cannot get the expected help.
- The bug has not been fixed in the latest version.
Describe the bug
I have been converting models to onnx, followed by trying to parse the models with a TensorRT version of onnx but the parsing fails. No error what's so ever, just return false. The steps for reproducing the error are below.
However, when parsing other onnx files, there are no issues. It happens only with mmdeploy version.
The odd part is, that I can open it with Netron, so it's not broken because of the "cp" bug in linux.
**Reproduction**
Weights:
wget -P checkpoints https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth
Model conversion:
python ~/mmdeploy/tools/deploy.py \ ~/mmdeploy/configs/mmdet/detection/detection_onnxruntime_dynamic.py \ ~/mmdetection/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py \ ~/checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \ ~/mmdetection/demo/demo.jpg \ --work-dir mmdeploy_model/faster-rcnn \ --device cpu \ --dump-info
So far looks alright, but when loaded with external onnx, the parsing fails. Pathing is right, since I check it. I am using latest TensorRT (8.4).
TensorRT - https://github.com/NVIDIA/TensorRT with their onnx - https://github.com/onnx/onnx-tensorrt/tree/9f82b2b6072be6c01f65306388e5c07621d3308f
auto parser = std::unique_ptr<nvonnxparser::IParser>(
nvonnxparser::createParser(*network, m_logger));
if (!parser) {
throw std::runtime_error("Parser failed");
}
std::ifstream file(onnxModelPath, std::ios::binary | std::ios::ate);
std::streamsize size = file.tellg();
file.seekg(0, std::ios::beg);
std::vector<char> buffer(size);
if (!file.read(buffer.data(), size)) {
throw std::runtime_error("Unable to read engine file");
}
auto parsed = parser->parse(buffer.data(), buffer.size());
if (!parsed) {
throw std::runtime_error("Parser failed");
}
- Did you make any modifications on the code or config? Did you understand what you have modified? No modifications were made.
Environment 2022-08-31 19:56:43,400 - mmdeploy - INFO - Environmental information 2022-08-31 19:56:43,574 - mmdeploy - INFO - sys.platform: linux 2022-08-31 19:56:43,574 - mmdeploy - INFO - Python: 3.8.13 (default, Mar 28 2022, 11:38:47) [GCC 7.5.0] 2022-08-31 19:56:43,574 - mmdeploy - INFO - CUDA available: True 2022-08-31 19:56:43,574 - mmdeploy - INFO - GPU 0: NVIDIA A10G 2022-08-31 19:56:43,574 - mmdeploy - INFO - CUDA_HOME: /usr 2022-08-31 19:56:43,574 - mmdeploy - INFO - NVCC: Cuda compilation tools, release 10.1, V10.1.24 2022-08-31 19:56:43,574 - mmdeploy - INFO - GCC: gcc (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0 2022-08-31 19:56:43,574 - mmdeploy - INFO - PyTorch: 1.11.0+cu113 2022-08-31 19:56:43,574 - mmdeploy - INFO - PyTorch compiling details: PyTorch built with:
- GCC 7.3
- C++ Version: 201402
- Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191122 for Intel(R) 64 architecture applications
- Intel(R) MKL-DNN v2.5.2 (Git Hash a9302535553c73243c632ad3c4c80beec3d19a1e)
- OpenMP 201511 (a.k.a. OpenMP 4.5)
- LAPACK is enabled (usually provided by MKL)
- NNPACK is enabled
- CPU capability usage: AVX2
- CUDA Runtime 11.3
- 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_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
- CuDNN 8.2
- Magma 2.5.2
- Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.3, CUDNN_VERSION=8.2.0, 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 -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -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-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=1.11.0, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=OFF, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF,
2022-08-31 19:56:43,574 - mmdeploy - INFO - TorchVision: 0.12.0+cu113 2022-08-31 19:56:43,574 - mmdeploy - INFO - OpenCV: 4.5.5 2022-08-31 19:56:43,574 - mmdeploy - INFO - MMCV: 1.6.0 2022-08-31 19:56:43,574 - mmdeploy - INFO - MMCV Compiler: GCC 9.3 2022-08-31 19:56:43,574 - mmdeploy - INFO - MMCV CUDA Compiler: 11.3 2022-08-31 19:56:43,574 - mmdeploy - INFO - MMDeploy: 0.7.0+83b11bc 2022-08-31 19:56:43,574 - mmdeploy - INFO -
2022-08-31 19:56:43,574 - mmdeploy - INFO - Backend information 2022-08-31 19:56:43,962 - mmdeploy - INFO - onnxruntime: 1.8.1 ops_is_avaliable : True 2022-08-31 19:56:43,986 - mmdeploy - INFO - tensorrt: 8.4.1.5 ops_is_avaliable : True 2022-08-31 19:56:44,000 - mmdeploy - INFO - ncnn: None ops_is_avaliable : False 2022-08-31 19:56:44,102 - mmdeploy - INFO - pplnn_is_avaliable: True 2022-08-31 19:56:44,117 - mmdeploy - INFO - openvino_is_avaliable: True 2022-08-31 19:56:44,131 - mmdeploy - INFO - snpe_is_available: False 2022-08-31 19:56:44,131 - mmdeploy - INFO -
2022-08-31 19:56:44,131 - mmdeploy - INFO - Codebase information 2022-08-31 19:56:44,132 - mmdeploy - INFO - mmdet: 2.25.0 2022-08-31 19:56:44,132 - mmdeploy - INFO - mmseg: None 2022-08-31 19:56:44,132 - mmdeploy - INFO - mmcls: None 2022-08-31 19:56:44,132 - mmdeploy - INFO - mmocr: None 2022-08-31 19:56:44,132 - mmdeploy - INFO - mmedit: None 2022-08-31 19:56:44,132 - mmdeploy - INFO - mmdet3d: None 2022-08-31 19:56:44,132 - mmdeploy - INFO - mmpose: None 2022-08-31 19:56:44,132 - mmdeploy - INFO - mmrotate: None
Error traceback None just fails.
Bug fix I have no clue why it happens unless you modified onnx with some customer stuff.
Warnings when making the onnx file.
[2022-08-31 19:27:47.396] [mmdeploy] [info] [model.cpp:95] Register 'DirectoryModel'
[2022-08-31 19:27:48.821] [mmdeploy] [info] [model.cpp:95] Register 'DirectoryModel'
/home/ubuntu/mmdetection/mmdet/datasets/utils.py:66: 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.
warnings.warn(
[2022-08-31 19:27:51.289] [mmdeploy] [info] [model.cpp:95] Register 'DirectoryModel'
2022-08-31 19:27:51,295 - mmdeploy - INFO - Start pipeline mmdeploy.apis.pytorch2onnx.torch2onnx in subprocess
load checkpoint from local path: /home/ubuntu/checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth
/home/ubuntu/mmdetection/mmdet/datasets/utils.py:66: 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.
warnings.warn(
2022-08-31 19:27:52,719 - mmdeploy - WARNING - DeprecationWarning: get_onnx_config will be deprecated in the future.
2022-08-31 19:27:52,719 - mmdeploy - INFO - Export PyTorch model to ONNX: mmdeploy_model/faster-rcnn/end2end.onnx.
/home/ubuntu/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)
/home/ubuntu/mmdetection/mmdet/models/dense_heads/anchor_head.py:123: UserWarning: DeprecationWarning: anchor_generator is deprecated, please use "prior_generator" instead
warnings.warn('DeprecationWarning: anchor_generator is deprecated, '
/home/ubuntu/mmdetection/mmdet/core/anchor/anchor_generator.py:333: UserWarning: grid_anchors would be deprecated soon. Please use grid_priors
warnings.warn('grid_anchors would be deprecated soon. '
/home/ubuntu/mmdetection/mmdet/core/anchor/anchor_generator.py:369: UserWarning: single_level_grid_anchors would be deprecated soon. Please use single_level_grid_priors
warnings.warn(
/home/ubuntu/mmdeploy/mmdeploy/codebase/mmdet/models/dense_heads/rpn_head.py:78: 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:]
/home/ubuntu/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)
/home/ubuntu/mmdeploy/mmdeploy/codebase/mmdet/core/bbox/delta_xywh_bbox_coder.py:39: 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(0) == bboxes.size(0)
/home/ubuntu/mmdeploy/mmdeploy/codebase/mmdet/core/bbox/delta_xywh_bbox_coder.py:41: 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) == bboxes.size(1)
/home/ubuntu/mmdeploy/mmdeploy/codebase/mmdet/deploy/utils.py:47: TracerWarning: Using len to get tensor shape might cause the trace to be incorrect. Recommended usage would be tensor.shape[0]. Passing a tensor of different shape might lead to errors or silently give incorrect results.
assert len(max_shape) == 2, 'max_shape should be [h, w]'
/home/ubuntu/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)
/home/ubuntu/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)
/home/ubuntu/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)
/home/ubuntu/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)
/home/ubuntu/anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/ops/nms.py:171: 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
/home/ubuntu/anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/ops/nms.py:172: 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)
/home/ubuntu/anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/ops/nms.py:31: 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:
/home/ubuntu/anaconda3/envs/openmmlab/lib/python3.8/site-packages/mmcv/ops/roi_align.py:83: 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 rois.size(1) == 5, 'RoI must be (idx, x1, y1, x2, y2)!'
/home/ubuntu/anaconda3/envs/openmmlab/lib/python3.8/site-packages/torch/onnx/symbolic_opset9.py:2905: 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.
warnings.warn("Exporting aten::index operator of advanced indexing in opset " +
2022-08-31 19:28:03,897 - mmdeploy - INFO - Execute onnx optimize passes.
2022-08-31 19:28:04,247 - mmdeploy - INFO - Finish pipeline mmdeploy.apis.pytorch2onnx.torch2onnx
2022-08-31 19:28:04,611 - mmdeploy - WARNING - "visualize_model" has been skipped may be because it's running on a headless device.
2022-08-31 19:28:04,611 - mmdeploy - INFO - All process success.
Just exported the model form here - https://github.com/WongKinYiu/yolov7. Works fine, but I need to add support for mmdet models.
-
If you want a .engine (TRT format) file, DEPLOY_CONFIG should directly use
configs/mmdet/detection/detection_tensorrt_dynamic-320x320-1344x1344.py, not first convert to .onnx then .engine -
onnx -> engine fails: trt not support the TOPK with dynamic K value, please search
topkfrom onnx file, check the input. If usingdetection_tensorrt_dynamicXXX, mmdeploy rewriter would rewrite the K value of topk to a fixed integer -
try this config to convert faster-rcnn:
✗ python3 tools/deploy.py configs/mmdet/detection/detection_tensorrt_dynamic-320x320-1344x1344.py ../mmdetection/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth ../mmdetection/demo/demo.jpg --work-dir faster-rcnn --device cuda:0 --show
If someone would like use cu102, please also cherry-pick this fix . Yes I have noticed that you are using cu113.