Description
Hi,
I am using CenterNet Resnet101 V1 FPN 512x512 [1]
model for detection and returned very good result for detection of objects. I wanted to accelerate my model with TensorRT 8.4 so I converted the “.pb” tensorflow files to .onnx model with tf2onnx tool. After extracted .onnx model, when I converted to TensorRT engine with trtexec (.trt), I am getting error. This error relevant with datatype, so I changed data format to float32 [2]. And then tried to updated onnx file to convert engine, I got error about padding [3]. So I saw solution for that issue and I applied but result not changed [4].
Also I’m using jetson orin 64gb with jetpack last version.
Can you
Environment
TensorRT Version: 8.4
GPU Type: Jetson Orin
Nvidia Driver Version: Tegra
CUDA Version: 11.4
CUDNN Version: 8.4
Operating System + Version: JEtson Orin
Python Version (if applicable): 3.8
TensorFlow Version (if applicable): 2.8
Relevant Files
[1] http://download.tensorflow.org/models/object_detection/tf2/20200711/centernet_resnet101_v1_fpn_512x512_coco17_tpu-8.tar.gz
[2] CenterNet keypoint detector not giving good FPS in jetson xavier NX - #5 by AastaLLL
[3] [E] Error[4]: [shuffleNode.cpp::symbolicExecute::390] Error Code 4: Internal Error (StatefulPartitionedCall/ResizeToRange/Pad/paddings_Unsqueeze__197: IShuffleLayer applied to shape tensor must have 0 or 1 reshape dimensions: dimensions were [1,2])
[4] IShuffleLayer applied to shape tensor must have 0 or 1 reshape dimensions: dimensions were [-1,2]) - #5 by spolisetty
Steps To Reproduce
Please include:
- Exact steps/commands to build your repro
- Exact steps/commands to run your repro
- Full traceback of errors encountered