This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

ARM compute library deconvolution validate() function boundary check

In ARM compute libarary 21.08 deconvolution validate() function has a boundary check, https://github.com/ARM-software/ComputeLibrary/blob/v21.08/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp

code portion:

const unsigned int out_x = (input->dimension(idx_w) - 1) * stride_x + 1;

ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(idx_w) > out_x);

Is this check correct? Since It violate for input weight as 1, weight width as 2 and stride as 2.

Parents
  • Questions on the Arm Compute Library (ACL) are best put as issues at the ACL github: https://github.com/ARM-software/ComputeLibrary/issues There the ACL experts will get back to you quickly.

    I'm not familiar with this bit of ACL, but from first look, it looks correct? its checking dimensions are compatible between weights, inputs and output, and I don't think your Stride is compatible with your Width for your error case? Also I'm not quite sure what you're referring to with the "input weight" with respect to this function?

Reply
  • Questions on the Arm Compute Library (ACL) are best put as issues at the ACL github: https://github.com/ARM-software/ComputeLibrary/issues There the ACL experts will get back to you quickly.

    I'm not familiar with this bit of ACL, but from first look, it looks correct? its checking dimensions are compatible between weights, inputs and output, and I don't think your Stride is compatible with your Width for your error case? Also I'm not quite sure what you're referring to with the "input weight" with respect to this function?

Children
No data