Hi,

I'd like to share a side project I worked on which generalizes transform coding to the nonlinear case. Here the transforms are represented by neural networks, which learn the appropriate form of the transform. The result of the transform is then
quantized using standard entropy coding.

Github: https://github.com/Justin-Tan/high-fidelity-generative-compression Interactive Demo: https://colab.research.google.com/github/Justin-Tan/high-fidelity-generative-compression/blob/master/assets/HiFIC_torch_colab_demo.ipynb

There are some obvious shortcomings to this method - such as, that it only caters for image data, cannot be adjusted to attain a variable bitrate, short of training a different model, and is unrealistically slow for practical applications.

I'm not a traditional compression expert, so would appreciate any insight about the deficiencies of this method from those who are. Note this is not my original idea and is a reimplementation.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Sun, 20 Sep 2020 21:29:48 -0700 (PDT), Justin Tan
<justin.tan@coepp.org.au> wrote:

Hi,

I'd like to share a side project I worked on which generalizes transform coding to the nonlinear case. Here the transforms are represented by neural networks, which learn the appropriate form of the transform. The result of the transform is then

quantized using standard entropy coding.

Github: https://github.com/Justin-Tan/high-fidelity-generative-compression >Interactive Demo: https://colab.research.google.com/github/Justin-Tan/high-fidelity-generative-compression/blob/master/assets/HiFIC_torch_colab_demo.ipynb

There are some obvious shortcomings to this method - such as, that it only caters for image data, cannot be adjusted to attain a variable bitrate, short of training a different model, and is unrealistically slow for practical applications.

I'm not a traditional compression expert, so would appreciate any insight about the deficiencies of this method from those who are. Note this is not my original idea and is a reimplementation.

EasyNN has image mode built in. I don't know how well it compresses
images because the person who tested and validated image encoding has
retired. I wrote the code a long time ago but I forget how it works.
I'm getting old!

Steve

--
http://www.npsnn.com

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

In comp.compression, Justin Tan <justin.tan@coepp.org.au> wrote:

I'd like to share a side project I worked on which generalizes transform coding to the nonlinear case. Here the transforms are represented by
neural networks, which learn the appropriate form of the transform. The result of the transform is then quantized using standard entropy coding.

Github: https://github.com/Justin-Tan/high-fidelity-generative-compression Interactive Demo: https://colab.research.google.com/github/Justin-Tan/high-fidelity-generative-compression/blob/master/assets/HiFIC_torch_colab_demo.ipynb

There are some obvious shortcomings to this method - such as, that it
only caters for image data, cannot be adjusted to attain a variable
bitrate, short of training a different model, and is unrealistically
slow for practical applications.

Also:

Clone repo and grab the model checkpoint (around 2 GB).

If you need 2GB of data around to compress / decompress images, you need
a lot of images before this starts "winning".

I'm not a traditional compression expert, so would appreciate any
insight about the deficiencies of this method from those who are. Note
this is not my original idea and is a reimplementation.

I'm no expert in compression, I just read this group for the occasional insight.

Elijah
------
particularly interested in image compression

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)