RAWerse Alpha

The web-server (and also web hooks) was down for a whole week. And I even haven't noticed that, only wondered "oh, 0 visits this day again :(".

Pro-tip: always check your website.

What actually happened - "Kernel panic" with no reason, haven't actually expected such thing using cloud instance, huh.

After setting up payment system and finished small license module, the RAWerse Alpha was published to the website. There are 2 version for users: Free Demo and Full Version.
Free demo, of course, has quite limited feature set: it can run only on CPU, there is no way to export DNG and image processing can be only done with default settings.

After research it took about a month to complete the application. The application itself contains 2 parts: back-end written in C++ and front-end written in JS with React / Electron.
Research and development were performed in Linux environment and it was a big pain to compile / port it to a target platform (Windows 64 bit).

In fact, it's not finished milestone, but I already got some good-enough results.

The main problems:

• it should work and work fast. It's the main problem with neural networks, especially pix2pix. It produces fast or best results. Pick one, never both.
• it should not make good images worse.
• it should not alter image colors / brightness.
• it should not make image blurry.

I have lots of images corrupted after JPEG compression (social media, old camera) and one day I realized, hey, I can fix these photos with the power of neural networks.

I tried to find any good (and affordable) application to accomplish that and then decided to make my own.