2d DCT Transform

[lifeblack (life)]

Hi!
I want to transform a tensor by 2d DCT transform… How could I do this using PyTorch that the output be as the same as matlab?

[Nikronic (Nikan Doosti)]

Hi,

I think this post can answer your questions:

[solved] Discrete cosine transform implementation in pytorch?

UPDATE: I figured it out. This link explains how to incorporate scipy functions into the computational graph. You can use scipy.fftpack.dct http://pytorch.org/tutorials/advanced/numpy_extensions_tutorial.html Hi everyone, So I want to backprop the error gradient through the inverse discrete cosine transform. I tried implementing a naive versi on of the DCT but it was reallyyyyy slow. Does anyone have any suggestions for writing a fast, backpropable implementation of the inverse discrete cos…

Bests

[lifeblack (life)]

thanks for your reply @ Nikronic… but it doesn’t work properly and its outputs are wrong (at least in my problem) …

[Nikronic (Nikan Doosti)]

What is your case? Can you create a toy example using numpy?
I have not worked with the code I referenced but at a glimpse, it seems alright.

[d4riush]

I developed a package that handles 2D DCT encoding/decoding:

GitHub - dariush-bahrami/dct-autoencoder: Discrete Cosine Transform in PyTorch