Gaussian grids speed up and shrink image data storage
Frequency-Adaptive Gaussian Grids for Efficient Image Representation
Computer Vision and Pattern RecognitionGraphics
Summary
Capturing high-resolution images in computers requires large amounts of memory and slow processing. The paper introduces Gaussian-Grid, a new way to represent images using a combination of structured grids and flexible 2D Gaussian shapes that adjust to details in the image. This hybrid approach chooses the best method automatically depending on how complicated each part of the image is, making storage smaller and decoding faster. The authors tested their method on natural images and showed it can be much faster and reduce data size much more than current common methods like JPEG with ZIP compression.
What this means in practice
- •For graphics software developers: Store and display high-resolution images with faster decoding and less memory using Gaussian-Grid representations on GPUs.
- •For video game engine teams: Reduce texture data transfer and speed up loading times by using frequency-adaptive Gaussian grids for image assets.
Authors
Anatolii Evdokimov, Stavros Diolatzis, Bingxuan Li, Akshay Jindal, Patsorn Sangkloy, Qi Sun
Abstract
Neural representation has emerged as a powerful tool to mitigate the demand for increasingly higher-resolution images. However, a fundamental trade-off remains between structured (e.g., feature grids, which offer spatial consistency) and unstructured (e.g., 2D Gaussians, which adaptively capture local variances) representations. This choice significantly affects how well an algorithm represents an image. Here, we present Gaussian-Grid, a hybrid neural representation that adaptively integrates structured and unstructured primitives. By dynamically partitioning computation between 2D Gaussian distributions and their attached multi-resolution feature grids based on the local frequency content, Gaussian-Grid automatically optimizes its architecture to match the underlying complexity of the target signal. We evaluate our approach using high-resolution natural images and demonstrate its applications as GPU-native dataset storage, showing $1.24$-$2.14\times$ decoding speed improvement and PCIe data traffic reduction of $23\times$-$176\times$ over JPEG + ZIP data archiving. The results show that Gaussian-Grid maintains small memory footprints while achieving high reconstruction fidelity across a variety of datasets.