Closed OuHangKresnik closed 5 years ago
I just checked the implementation of this operator in tensorflow, https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/kernels/strided_slice.cc
the logic looks more complex than it should be.
I will try to replace this operator with constant, while since there are dozens of them (maybe one hundred), I don't know whether this would work. Another thing is, this operator contains lots of 0 dim output. As long as TASO support 0-dim, then shouldn't be a blocker
Actually 2 situations:
These 2 are different.
It's just very hard to replace this op with constant, since there are hundreds of them and need to replace both shape and value. (If value only, the following operators would result in wrong shapes)
When I tried to dump the real value, there is a bug from tensorflow DebugString() which causes segmentation fault
Just as this error shows