Input image processing in TensorRT 8-bit mode

We know that the concrete value of input image is range from 0 to 255. After mean value subtraction, the range is changed to -255 to 255, which should be represented by a int9 value. Does tensorrt 8-bit mode have some special processing for the input layer or just convert input image into FP32_scale_factor*int8 array? I thought the latter one will cause information loss of input image.