Open doruksonmez opened 2 years ago
TensorRT does not support the Round
operation yet. TRT will attempt to load user-defined plugins if an operator is not supported, which is why the following error message appears.
[E] [TRT] 3: getPluginCreator could not find plugin: Round version: 1
The operator will be supported in a future release
Description
Hi all. I am currently trying to convert Faster RCNN models from TF2 Model Zoo from saved model to TensorRT engine. It is mainly related to
Unsupported ONNX data type: UINT8(2)
error but I solved it using the following snippet from ELinux: Unsupported ONNX data type: UINT8(2) Fix.Now I'm having another issue after solving it when I try to convert fixed ONNX model to TensorRT engine using
trtexec
. I get the following errors:Environment
TensorRT Version: 8.0.3 ONNX-TensorRT Version / Branch: GPU Type: RTX 3090 Nvidia Driver Version: 470 CUDA Version: 11.3 CUDNN Version: 8.2.4 Operating System + Version: Ubuntu 20.04 Python Version (if applicable): 3.8 TensorFlow + TF2ONNX Version (if applicable): TF 2.6.0 + TF2ONN 1.9.3/1190aa PyTorch Version (if applicable): Baremetal or Container (if container which image + tag): nvcr.io/nvidia/tensorflow:21.10-tf2-py3
Relevant Files
Steps To Reproduce
From SavedModel to ONNX
python3 -m tf2onnx.convert --saved-model faster_rcnn_resnet50_v1_800x1333_coco17_gpu-8/saved_model --output faster_rcnn_resnet50_v1_800x133_trt.onnx --opset 11 --signature_def serving_default --tag serve --target tensorrt
From ONNX to TensorRT
trtexec --onnx=/root/tf2_models/faster_rcnn_resnet50_v1_800x1333_trt/faster_rcnn_resnet50_v1_800x133_trt.onnx --fp16 --buildOnly --saveEngine=/root/tf2_models/faster_rcnn_resnet50_v1_800x1333_trt/faster_rcnn_resnet50_v1_800x133_trt.trt