I’m doing image processing in the frequency domain using CUFFT to transform the data. I’m seeing artifacts from the FFT segmentation in the 3.1 toolkit. In the output, the images appear gridded up in a factor of two. The various grids have different intensities from one another. This would not be as big of an issue but a visible “barrier” is seen at the edges of the segments if you plot the image in 3D. I’ve seen this same type of artifact in images with dimensions that are multiples of 2, 3, and 5.

Do you use in-place or out-of-place transformation? When I use in-place transform in 3.1, I had strange periodic error - corrupted samples with period equal approximately FFT size. I realized out-of-place transform, and result become correct.

Do you use in-place or out-of-place transformation? When I use in-place transform in 3.1, I had strange periodic error - corrupted samples with period equal approximately FFT size. I realized out-of-place transform, and result become correct.

I used a combination of in- and out-of-place transforms originally. I simplified the code to perform only a real-to-complex and complex-to-real out-of-place transform on the original image. The expected result is the original image, but I get that same periodic issue as before.

I used a combination of in- and out-of-place transforms originally. I simplified the code to perform only a real-to-complex and complex-to-real out-of-place transform on the original image. The expected result is the original image, but I get that same periodic issue as before.