Hi everyone! I just trained an image classification model on my PC using tensorflow, resulting in an .pb model. I converted it into onnx and used it on my jetson with jetson-inference. But surprisingly the results differ from what i get when i use inference on my PC (with either tensorflow or onn…

Hi @AastaLLL , i managed to run TensorRT inference by adapting your script to the changes in TensorRT API 8.0.1. I changed the image preprocessing in tensorflow and TensorRT to match the preprocessing in training and now i get same predictions on both. Here is the script i used for TensorRT infere…

Jetson-Inference predictions differ from e.g. tensorflow predictions

Autonomous Machines Jetson & Embedded Systems Jetson Nano

AastaLLL October 25, 2021, 2:13am 3

Hi,

It will be good to check this issue is from TensorRT or jetson_inference itself.
Would you mind running your model with pure TensorRT directly?

You can find an example below:

Thanks.