# Trying to build a DLSS-like remastering model for GBA games — retained record

> Retained reddit record from reddit:computervision.

Canonical: https://highsignal.app/data/records/b5d63cc2a333b4be

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:computervision)  published 2026-09-26 00:50 ingested 2026-09-27 02:36

[original source ↗](https://www.reddit.com/r/computervision/comments/1wqcyil/trying_to_build_a_dlsslike_remastering_model_for/)

## retained text



```
I’m experimenting with a DLSS-like system for GBA games.

The goal isn’t simple upscaling. I want a model to take a 240×160 frame and reconstruct it into a higher-resolution remastered version while keeping the original layout and style.

The core idea is to reuse the previous high-resolution frame as the base, then only modify the parts that changed:

Output = Base + Gate × Residual

For training data, I’m planning to use my own game engine. I can generate random maps and render the exact same scene twice:

low-resolution retro version

high-resolution remastered version

This gives me perfectly aligned LR/HR training pairs automatically. I can also export object masks, motion data, and other metadata if needed.

I’ve trained and deployed several deep learning models in production before, but I don’t have much experience designing and training a model architecture from scratch. This one is also image-to-image, so I’m not yet confident the approach will work as well as I’m imagining.

If anyone here has experience with image restoration, video super-resolution, temporal consistency, or neural rendering, I’d really appreciate feedback on the architecture and training approach.
```



document fetched 2026-09-27 00:23 · canonical https://reddit.com/r/computervision/comments/1wqcyil/trying_to_build_a_dlsslike_remastering_model_for
