Transforming a rtr ver 8.5.5.2 yolox model onnx / pth file into a 6.2 jetpack (super mode) fit networm

Robotics & Edge Computing Jetson & Embedded Systems Jetson Orin Nano
,
1

Hi,
I can’t work with the .engine file produced for older rtr 8.5.5.2 created with mmdeploy
I upgraded to jetpack 6.2 and now i cant work and cant transform it to a fitted .engine file..

now im at 10.3.0.30 trt and cuda 12.5
what versions of pytorch, mmdeplot, mmcv and all other i need
and how do i install all of them because there are very problematic

5

thanks,
the mmdeploy has lots of mismatches lik in trt_bached_nms.cpp files having int instead of int64 (fit to trt10) do you have someone who already patched all of these files?

6

Hi,

Based on their document, they need TensorRT 8 to work:

You can check with with MMDeploy team to see if they are going to support TensorRT 10.

Thanks.

7

do you thing TensorRT 8 can be installed on a jetson 6.2 ( super mode)

8

Hi,

Due to GPU driver dependencies, it might not work if installed natively.
If you want to try, you can test it with the JetPack 6.0 container:

Thanks.

9

ok, but here you just open a 6.2 jetpack fit docker.. and dont solve the problem of the mmdeploy right?

10

Hi,

The container uses JetPack 6.0 which has TensorRT 8.6.
If mmdeploy can work on TensorRT 8, it should be able to work on that container.

Thanks.