FEATURE REQUEST Be compatible with UINT8 input data type

[dmenig]

TensorRT can't take UINT8 inputs, unlike openvino or pytorch for example. This topic makes me think it's pretty important to users though onnx/onnx-tensorrt#400

This means that all conversions must be done on CPU before feeding the data to GPU.

I think it'd be useful if TensorRT could take UINT8 inputs, since it's the most common data type for image data.

[nvpohanh]

UINT8 support is on roadmap.

[justinblaber]

This is especially useful if you want to do real time processing on large images, so would be nice if it was supported directly.

[ttyio]

I will close this issue since now UINT8 input is supported start TRT8.5, thanks all!

https://docs.nvidia.com/deeplearning/tensorrt/release-notes/index.html#rel-8-5-1