Decrease in accuracy of TensorRt model on jetson TX2

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

Hi,
I am trying to detect Iicense plate in veideos with yolov3 model. I inference TensorRt model(onnx->TensorRT) on jetson nano. And I found that some of the license plates can not be detected (which can be detected with onnx model). Is it normal ? And, is there a tool to compare the outputs of the onnx and tensorrt model ?

TensorRT Version : 7.1.0
Device : jetson TX2
CUDA Version : 10.2

3

Hi,

Suppose you are using darknet as the training framework.
Have you feed the same input video to the model before?

It will be good to give it a try.
This can help us to figure out this is a model issue or conversion issue.

Thanks.

4

HI ,

I have already feed the same input (both video and images) to onnx and tensorrt. And the outputs are slightly different which cause some of the license plates were not detected with the tensorrt version.

5

There is no update from you for a period, assuming this is not an issue any more.
Hence we are closing this topic. If need further support, please open a new one.
Thanks

Hi,

Could you share a simple reproducible source and the onnx model for us debugging?
We want to reproduce this issue internally before giving a further suggestion.

Thanks.