Support for dynamic batch size in scorer ONNX model

[ywlng]

Hi, thank you for sharing this project.
I found that the provided scorer_hwc.onnx declares a dynamic batch dimension, but internally it contains a hard-coded batch size of 252, so it cannot run with arbitrary batch sizes.
Could you please explain how scorer_hwc.onnx was exported? Would it be possible to replace it with the official NVIDIA FoundationPose score_model.onnx from NGC and build the TensorRT engine for this project?
My use case requires a truly dynamic batch size. Any guidance on the required model conversion or input/output changes would be appreciated.

[zz990099]

Hello~
Using the model from NGC and converting it to TensorRT should be feasible. If the batch size is kept at 252, it might work directly with the current codebase. However, if you need to change the batch size to other values, the existing code does not currently support that.