Cannot inference with deployed TensorRT engine on Jetson AGX Xavier.

[tung18tht]

Hi, I am trying to setup the mmdeploy on my Jetson AGX Xavier. Everything went fine including the conversion step.
I converted the yolox-s detection model and got the end2end.engine as the output.

But when I try to inference with the command ./build/bin/object_detection cuda ./end2end.engine ./demo.jpg, it has the following error:

[2022-12-10 14:01:29.475] [mmdeploy] [error] [model.cpp:39] Failed to load model: "./end2end.engine", implementations tried: [("DirectoryModel", 0)]
[2022-12-10 14:01:29.475] [mmdeploy] [error] [model.cpp:16] Failed to load model "./end2end.engine"
[2022-12-10 14:01:29.475] [mmdeploy] [error] [model.cpp:21] failed to create model: not supported (2) @ .../mmdeploy/csrc/mmdeploy/core/model.cpp:41
failed to create detector, code: 6

I already tried reinstalling, and using python API, but the problem persists. Do you guys have any idea on the problem and how to fix it?

[lvhan028]

Please read https://github.com/open-mmlab/mmdeploy/blob/master/docs/en/get_started.md

build/bin/object_detection is implemented on MMDeploy Inference SDK.
SDK performs model inference with not only the backend engine file but also meta info.

How to get meta info?
Use --dump-info when converting an OpenMMLab model by tools/deploy.py.
The generated meta info, as well as the backend engine file, are located in the directory specified by --work-dir
They make up what we call MMDeploy Model.

After exporting MMDeploy Model, pass the path specified by --work-dir to build/bin/object_detection