sreevenkjan
Full Member level 5
2D defines the input data size. Its 2 dimensional. It depends on the kernel size. Like I said you need to do FFT one time in the x-direction, save the data into the buffers and then once the data are saved in buffers then you have 3 rows of data (assuming you have 4x4 kernel). Then once you have the data in buffers, you do convolution in y-direction. To verify your results, I suggest you collect the FFT data and check it with your simulated output in Matlab. This way you can say if you are right or wrong.