MNN2.0 conversion reasoning result is incorrect

[Ley-WL]

Platform(Include target platform as well if cross-compiling): ubuntu1604, gcc version 5.4.0 20160609

Github Version: MNN2.0

Provide date (or better yet, git revision from the comment section of the zip. Obtainable using 7z l PATH/TO/ZIP and search for Comment in the output) if downloading source as zip,otherwise provide the first commit id from the output of git commit Comment = 4679f848c45510531976ebdf32c42b1c27b92960

Compiling Method

GCC G++

Paste cmake arguments or path of the build script used here as well as the full log of the cmake proess here or pastebin cd MNN/ ./schema/generate.sh mkdir build cd build cmake .. -DMNN_BUILD_CONVERTER=true && make -j4

Problem: The model transformed by the 2.0 compiled transformation tool cannot be matched with the reasoning result of 2.0. Cross verify the two versions. If 1.1 reasoning and 2.0 transformation results are selected, the 1.1 transformation model is selected, and the reasoning results of 2.0 and 1.1 are completely consistent and correct.

[jxt1234]

See https://www.yuque.com/mnn/cn/model_convert#nxImR. Use fastTestOnnx.py / fastTestTf.py to check if MNN's result is the same as TF / ONNX