Hello, thanks for providing the robust testing tool.
Fourier transform torch.fft.fft is used in my model (Complex numbers are extracted into two channels [tensor.real, tensor.imag]).
Attacks can be carried out sequentially in stages apgd-ce, apgd-t, and fab-t.
But when conducting a square attack, an error occurs. cuFFT error: CUFFT_INVALID_SIZE.
I am confused whether autoattack does not support the Fourier transform.
Hello, thanks for providing the robust testing tool. Fourier transform
torch.fft.fft
is used in my model (Complex numbers are extracted into two channels[tensor.real, tensor.imag]
). Attacks can be carried out sequentially in stagesapgd-ce
,apgd-t
, andfab-t
. But when conducting a square attack, an error occurs.cuFFT error: CUFFT_INVALID_SIZE
. I am confused whether autoattack does not support the Fourier transform.