Closed
Description
fp16测试正常，但是量化成int8后没有预测框（测试了很多图片都没有框），您帮忙看看我结果是正常的量化现象，还是我哪个步骤出错了呢？？？
步骤1：生成imagelist.txt 200张
find val/ -type f > imagelist.txt
步骤2：生成.table文件
ncnn2table v5lites-hive-best-opt.param v5lites-hive-best-opt.bin imagelist.txt v5lites-hive-best.table mean=[0,0,0] norm=[0.003922,0.003922,0.003922] shape=[320,320,3] pixel=BGR thread=8 method=kl
步骤3：量化：fp16->int8
ncnn2int8 v5lites-hive-best-opt.param v5lites-hive-best-opt.bin v5lites-hive-best-int8.param v5lites-hive-best-int8.bin v5lites-hive-best.table
预测效果（左边fp16 右边int8）：
Metadata
Assignees
Labels
No labels