The calculation result is incorrect after updating the library (v1.6 -> v2.2)
Zhao-Dongyu opened this issue · 0 comments
Zhao-Dongyu commented
After I upgraded the version from 1.6 to 2.2, there was a difference in the calculation results of the convolution. The calculation result of the original 1.6 version is normal, but the result of the 2.2 version is wrong.
The wrong operator is Standard 2D Convolution Kernels.
Input shape is 1x9x80x1,
filter<80x3x3x1>,
dilation_h = 1, dilation_w = 1,
padding = SAME,
stride_h = 2, stride_w = 2,
Output shape is 1x5x40x80
I see that there is no change in the relevant documents, and the method of using the function is the same as before, but the calculation result is indeed wrong.
I'm a bit confused where the problem is :)