Deploying without or less dependency with MMdeploy, MMCV, MMEngine, etc.

[RiverLight4]

Hello,

I'd like to convert MMDetection trained model into other acceleration hardware platform.
But I'm afraid that some inference engine cannot use mmdeploy.dll or libmmdeploy_tensorrt_ops.so, mmdeploy_runtime, even if mmcv python modules, when inference is running on the hardware.

Here is my question:

• Is there any method to do inference with OpenMMLab converted model WITHOUT using OpenMMLab libraries or resources?
• If not, what libraries do I have to load?

Additional Info:

• I'm planning to do inference with TensorRT model on Jetson, Hailo model on Hailo-8, OpenVINO model on Intel CPU.
• I'm converting Mask R-CNN model, so I think libmmdeploy_tensorrt_ops.so or libmmdeploy_onnx_ops.so is always needed.