Request to Trace Grounded DINO Model into TorchScript for Deployment

[leleZeng]

Hi Grounded DINO team,

First of all, thank you for your incredible work on Grounded DINO! The model has been extremely useful in my research and development.

I would like to request support for converting the Grounded DINO model into a TorchScript format via tracing, so it can be deployed in environments that do not require Python. Specifically, this would allow us to:

Deploy the model in C++\JAVA or other non-Python environments: TorchScript models can be loaded and run in C++ or embedded systems without needing Python, making deployment more flexible.
Improve inference performance: TorchScript can optimize the model for faster execution by generating a static computation graph, which is often more efficient than the dynamic computation graphs of PyTorch.
Enable wider use cases: Some production environments or edge devices cannot rely on Python runtimes, and a TorchScript version would enable using Grounded DINO in such scenarios.
Proposal:
Could you provide a version of the model traced into TorchScript, using torch.jit.trace or torch.jit.script where applicable?
If possible, include a guide or example for how to trace the model ourselves for custom use cases.
This feature would greatly help those of us who are looking to deploy Grounded DINO in environments like C++\JAVA applications, edge devices, or production servers where Python is not feasible.

Looking forward to your response!