Resolution Enhancement

Super Resolution Enhancement for Large Displays & Printed Materials

Project Overview:

Many firms struggle to generate super high resolution images intended for large displays and printings. The traditional method of resolution enhancement takes tremendous amounts of time and cost, as well as being almost impossible without dedicated software and hardware. Planby offers AI-based super upscaling to handle this issue.

Key highlights:

Tackling the excessive resources required to upscale images to super high resolutions for large displays or large printed materials

Planby developed an AI resolution enhancer model capable of enhancing resolutions to super resolution levels within minutes, without the need for special hardware or software.

The AI resolution enhancer model was not only faster, cheaper and less resource consuming, but it also provided a better resolution to file size ratio.
super resolution detail image1

For large construction or real estate development projects, it is almost necessary to include a wide range perspective of the entire scenery like the image seen above. The problem arises when these images need to be presented on a large screen for a large audience, or when it needs to be printed at physically large scale like on construction site banners. Normal images, when enlarged to such large sizes, show blurs and pixelations in the details of the image, making the images far less aesthetically pleasing. So super resolution upscaling is almost necessary in such situations, but traditional upscaling procedures require tremendous resources and dedicated software & hardware.

super resolution detail image2

So the Planby team developed a spatial design focused AI resolution enhancer model to resolve this issue. There are many upscale AI models readily available, but none of them quite meet the standards of this specific situation. These occasions require super resolution beyond 4K, which is the typical limit for generic upscale AI models. In addition, some upscale AI models cannot maintain the level of detail while upscaling, which defeats the whole purpose of upscaling. Our team experimented with countless upscale AI models, compared the results and fine-tuned the best upscale AI model by balancing the level of upscaling and the loss of detail to optimize for wide range perspective images.

super resolution detail image3

The image above shows the comparison between the original image and the enhanced image. It is apparent that the upscaled image is much more clear with the smallest details preserved. Although unintended, the enhanced image also had a much smaller size (tens of megabytes) compared to the size of traditionally upscaled images (up to gigabytes), adding another bonus to the AI super resolution upscale model.

Results and Effects:

Successful demonstration of how generic AI models could be applied to domain-specific purposes with the AI & domain expertise of Planby.

Unexpected benefit of having a much smaller size, on top of being cheaper, easier and faster than the traditional method.

What solutions have other companies used?