Researchers at NTU Singapore Propose a Novel and Efficient Diffusion Model for Image Restoration IR that Significantly Reduces the Required Number of Diffusion Steps

Image restoration (IR) is a crucial task in computer vision, seeking to recover high-quality images from their degraded versions. Whether it’s an old, faded photograph or a photo blurred by camera shake, we want to fix these imperfections. Traditional methods have progressed, but diffusion models have recently emerged as a powerful solution for image restoration. However, existing diffusion models often need many steps to produce good results, slowing the restoration process.

Researchers have now developed a new diffusion model (as shown in Figure 2) specifically designed to make the image restoration process faster and more effective. It all starts with a simple insight: why start image restoration from scratch (like with random noise) when we already have a degraded version of the image? Their model smartly uses the degraded image as a basis for restoring the original, high-quality version.

The beauty of this new diffusion model (ResShift) lies in how it cleverly shifts the difference (or residual) between the degraded and original images. This approach allows them to use fewer steps while achieving excellent results. If you’re curious about the technical details, the model uses a carefully designed transition kernel and flexible noise schedule to control the image transformation process (as illustrated in Figure 3).

The researchers tested their model on various tasks like image super-resolution (making images sharper) and inpainting (filling in missing parts of images). The results were impressive! Their model was significantly faster than existing methods and often produced images that looked better to the human eye. For example, their model achieved impressive results in image super-resolution with just a handful of steps as shown in Figure 1. Imagine taking a blurry photo and making it crisp in record time! This opens up possibilities for real-time image restoration in cameras or photo editing software.

Our reflection on this study underscores the model’s innovative approach to balancing efficiency with performance, setting a new benchmark in the IR domain. Importantly, this model’s practical applications extend beyond the academic realm, with potential uses in real-time image restoration in cameras or photo editing software. However, it’s worth noting that further exploration is needed to fully understand the model’s limitations and potential in broader applications.

Check out the Paper and Github. All credit for this research goes to the researchers of this project. Also, don’t forget to follow us on Twitter. Join our Telegram Channel, Discord Channel, and LinkedIn Group.

If you like our work, you will love our newsletter..

Don’t Forget to join our 38k+ ML SubReddit

Want to get in front of 1.5 Million AI enthusiasts? Work with us here