Instant GaussianImage: A Generalizable and Self-Adaptive Image Representation via 2D Gaussian Splatting

2025-06-29

AI Summary

Existing implicit neural representations (INRs) suffer from high computational overhead, while GaussianImage reduces resource requirements but is constrained by a fixed number of Gaussians and slow convergence. To address these limitations, we propose an adaptive 2D Gaussian lattice-based image representation framework: a lightweight neural network rapidly generates an initial Gaussian distribution, and an image-entropy-aware mechanism dynamically modulates Gaussian density. Furthermore, we introduce deep-network-inspired initialization and efficient fine-tuning strategies to achieve high-fidelity rendering without compromising training speed. Experiments on DIV2K and Kodak demonstrate that our method achieves higher PSNR and SSIM than GaussianImage under identical Gaussian counts, reduces required training iterations by over 85%, and cuts runtime by an order of magnitude. To the best of our knowledge, this is the first approach to simultaneously achieve high reconstruction fidelity and unprecedented training efficiency in Gaussian-based image representation.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Representation Learning
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Implicit Neural Representation (INR) has demonstrated remarkable advances in the field of image representation but demands substantial GPU resources. GaussianImage recently pioneered the use of Gaussian Splatting to mitigate this cost, however, the slow training process limits its practicality, and the fixed number of Gaussians per image limits its adaptability to varying information entropy. To address these issues, we propose in this paper a generalizable and self-adaptive image representation framework based on 2D Gaussian Splatting. Our method employs a network to quickly generate a coarse Gaussian representation, followed by minimal fine-tuning steps, achieving comparable rendering quality of GaussianImage while significantly reducing training time. Moreover, our approach dynamically adjusts the number of Gaussian points based on image complexity to further enhance flexibility and efficiency in practice. Experiments on DIV2K and Kodak datasets show that our method matches or exceeds GaussianImage's rendering performance with far fewer iterations and shorter training times. Specifically, our method reduces the training time by up to one order of magnitude while achieving superior rendering performance with the same number of Gaussians.

Problem

• Reduce GPU resource demand in image representation
• Accelerate slow training process of Gaussian Splatting
• Adapt Gaussian point count to image complexity

Innovation

• Network generates coarse Gaussian representation quickly
• Minimal fine-tuning steps for high rendering quality
• Dynamic Gaussian point adjustment for image complexity