Thanks for your excellent work on the ONNX and TensorRT implementations.
I've noticed that the input size of the *_lightglue_pipeline.trt.onnx model is (2, 1, 1024, 1024). I'd like to adjust this to (2, 1, 512, 512).
I've found a conversion script for the *_lightglue_pipeline.onnx model, but I'm unsure how to modify it for the TRT model.
Could you please provide a script or guidance on how to achieve this input size adjustment?