[Feature/Bug] TensorRT backend detects .engine files but triggers NotImplementedError

[alex-itico]

The Bug

When trying to load a compiled TensorRT .engine model using the CLI or a configuration file, the wrapper successfully auto-detects the format as ModelType.TENSORRT. However, the execution crashes instantly because TensorRTBackend raises a NotImplementedError: TensorRT support not implemented yet. inside tensorrt_backend.py.

Environment Context

• Hardware: NVIDIA Jetson Orin Nano
• OS: Ubuntu 22.04 LTS (JetPack 6.x)
• Python Version: 3.10.12
• Library Version: Latest (installed via uv)

Steps to Reproduce

1. Compile a PaDiM model to TensorRT via trtexec:

   trtexec --onnx=model.onnx --saveEngine=model.engine --fp16
   

2. Configure configDetect.yml to point to the generated file:

model: "/path/to/model.engine"
device: "cuda"

3. Run detection:

anomavision detect --config configDetect.yml --thresh 12

[github-actions]

👋 Hello @alex-itico, thank you for opening this issue! Welcome to the AnomaVision community.

Our triage bot will route this to the maintainers for review.

Please hold tight while we route this to the right team member. 🚀

[DeepKnowledge1]

The wrapper Tensorrt mignt need re-implementstion

I will have a look

[DeepKnowledge1]

If you check the tensorrt_backend.py , all functions are NotImplementedError("TensorRT

You need add a correct code

[DeepKnowledge1]

@alex-itico

Try to adapt this implementation, if it is working, please make a pull request

tensorrt_backend.py

inference/model/backends/tensorrt_backend.py

[DeepKnowledge1]

@alex-itico , did you fix it