量化工具报错input tensor error错误

[haima1998]

我尝试用MNN量化，会报以下错误，请问这个问题大概是什么原因引起的？
MNN/tools/quantization/calibration.cpp:189: Check failed: inputTensorStatistic != _featureInfo.end() ==> input tensor error!

我的配置文件这么来写 preprocessConfig.json
{
"format":"RGB",
"mean":[
127.5,
127.5,
127.5
],
"normal":[
0.00784314,
0.00784314,
0.00784314
],
"width":544,
"height":544,
"path":"/path/images",
"used_image_num":500,
"feature_quantize_method":"KL",
"weight_quantize_method":"MAX_ABS"
}

如果把feature_quantize_method和weight_quantize_method改为ADMM，"input tensor error!" 问题就没有了，但是执行到最好会crash，crash在

#0 __memmove_avx_unaligned_erms () at ../sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:249
249 ../sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S: No such file or directory.
[Current thread is 1 (Thread 0x7f981f60b740 (LWP 22608))]
(gdb) bt
#0 __memmove_avx_unaligned_erms () at ../sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:249
#1 0x00007f981f18519a in MNN::CPUReshape::onExecute(std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&) () from /home/charlie/disk1/code/opensouce/MNN/build/libMNN.so
#2 0x00007f981f109c4e in MNN::Pipeline::executeCallBack(std::function<bool (std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, MNN::OperatorInfo const*)> const&, std::function<bool (std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, MNN::OperatorInfo const*)> const&) () from /home/charlie/disk1/code/opensouce/MNN/build/libMNN.so
#3 0x00007f981f116957 in MNN::Session::runWithCallBack(std::function<bool (std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, MNN::OperatorInfo const*)> const&, std::function<bool (std::vector<MNN::Tensor*, std::allocatorMNN::Tensor* > const&, MNN::OperatorInfo const*)> const&, bool) const () from /home/charlie/disk1/code/opensouce/MNN/build/libMNN.so
#4 0x000055f9fef1004d in Calibration::_computeFeatureScaleADMM() ()
#5 0x000055f9fef163e8 in Calibration::runQuantizeModel() ()
#6 0x000055f9feeeadaa in main ()
(gdb)

[czy2014hust]

1. MNN/tools/quantization/calibration.cpp

   Line 181 in cfad1c3

   if (_featureQuantizeMethod == "KL") {

   可以先把此if块注释掉
2. 如果出现这个错误的话，ADMM目前暂不支持

[haima1998]

1. MNN/tools/quantization/calibration.cpp

   Line 181 in cfad1c3

   if (_featureQuantizeMethod == "KL") {

   
   可以先把此if块注释掉
2. 如果出现这个错误的话，ADMM目前暂不支持

多谢了！ 注释掉if块后问题解决，MNN模型量化完成。
但是量化后的推理速度并未有明显提升，只是模型大小减小明显。