yolov5-face
yolov5-face copied to clipboard
About PTQ int8 optimization
Hi, guys,
Is that possible to apply the ptq int8 optimizations on yolov5face's torch2trt?
I tried to add the supports of int8 optimization by using widerface val-dataset for calibrations and compare it with fp16-only optimization, but found nearly-zero performance gains. Is my result correct?
Any help would be appreciated!