NVIDIA® TensorRT™ is an SDK for high-performance deep learning inference on NVIDIA GPUs. This repository contains the open source components of TensorRT.
Hi, we get a problem where a torch model only support image shapes which are multiple of 16, and the input shape is dynamic. But tensorrt dynamic shape only support min, opt and max, and it seems to have a step-size of 1 in the range of [min,max], which would cause the illegal input for my network. I wonder how to deal with it, i search the nvidia document and google, no answer.
Any suggestions? Thanks a lot
Hi, we get a problem where a torch model only support image shapes which are multiple of 16, and the input shape is dynamic. But tensorrt dynamic shape only support min, opt and max, and it seems to have a step-size of 1 in the range of [min,max], which would cause the illegal input for my network. I wonder how to deal with it, i search the nvidia document and google, no answer. Any suggestions? Thanks a lot