Hi guys, I still didn't understand the meaning of precision model FP32-INT8, for the example in this model https://download.01.org/opencv/2019/open_model_zoo/R4/20200117_150000_models_bin/head-pose-estimation-adas-0001/
Is the FP32-INT8 precision model support the range of FP32, FP16, and INT8? Or it was just support FP32 or INT8 ?
FP32-INT8 supposed to be a mixed precision model which means that it contains both FP32 and INT8 units/operations to increase the performance or decrease the memory usage and partially keep the accuracy value.
Hope this helps.
Best regards, Max.