ncnn-assets icon indicating copy to clipboard operation
ncnn-assets copied to clipboard

ncnn2int8量化时出现“Floating point exception (core dumped)”求助

Open viethung2002 opened this issue 6 months ago • 0 comments

内容: 大家好, 我在使用ncnn对YOLO模型进行量化时,执行以下命令:

./ncnn2int8 yolo11n.ncnn-opt.param yolo11n.ncnn-opt.bin yolo11n-int8.param yolo11n-int8.bin yolo11n.ncnn.table

运行到某些卷积层时,出现报错:

quantize_convolution conv_80

... quantize_convolutiondepthwise convdw_174 quantize_convolutiondepthwise convdw_175 quantize_convolutiondepthwise convdw_176 quantize_convolutiondepthwise convdw_177 quantize_convolutiondepthwise convdw_178 quantize_convolutiondepthwise convdw_179 quantize_convolutiondepthwise convdw_180 Floating point exception (core dumped)

我已经检查过table文件和param/bin文件都是有效的。 不清楚为什么会出现floating point错误。 请问有谁对ncnn2int8量化有经验,能帮我分析原因并给出解决方法吗?

非常感谢!

viethung2002 avatar Sep 02 '25 02:09 viethung2002