Description
When my model has INT8 weights should I pre-process input images from [0-255] to [0-1] range? I ask this because if images is in [0-1] I basically multiply floats (input) with integers (weights) and I fear this may reduce quantization benefit.
Thank you.