Intel Caffe int8 推理校准工具 统计各层输入数据参数 和 卷积参数 绝对最大值 计算量化尺度 This convert tools is base on TensorRT 2.0 Int8 calibration tools,which use the KL algorithm to find the suitable threshold to quantize the activions from Float32 to Int8(-128 - 127). - View it on GitHub