Generative Latent Diffusion for Efficient Spatiotemporal Data Reduction

2025-07-02

AI Summary

Generative models, despite their conditional modeling capability, suffer from poor controllability and low reconstruction fidelity, limiting their direct application to efficient spatiotemporal data compression. This paper proposes a generative latent diffusion compression framework that jointly leverages variational autoencoders (VAEs) and conditional diffusion models: only key frames are encoded into a compact latent space, and intermediate frames are synthesized via diffusion conditioned on these latent representations—enabling end-to-end, controllable interpolation-based reconstruction. By avoiding per-frame encoding, the method significantly improves both compression ratio and reconstruction fidelity. Evaluated on multiple spatiotemporal datasets, it achieves up to a 10× higher compression ratio than traditional methods (e.g., SZ3); under identical reconstruction error constraints, it outperforms state-of-the-art learning-based approaches by 63% in compression efficiency.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Data Mining & Knowledge Management: Data Compression

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Generative models have demonstrated strong performance in conditional settings and can be viewed as a form of data compression, where the condition serves as a compact representation. However, their limited controllability and reconstruction accuracy restrict their practical application to data compression. In this work, we propose an efficient latent diffusion framework that bridges this gap by combining a variational autoencoder with a conditional diffusion model. Our method compresses only a small number of keyframes into latent space and uses them as conditioning inputs to reconstruct the remaining frames via generative interpolation, eliminating the need to store latent representations for every frame. This approach enables accurate spatiotemporal reconstruction while significantly reducing storage costs. Experimental results across multiple datasets show that our method achieves up to 10 times higher compression ratios than rule-based state-of-the-art compressors such as SZ3, and up to 63 percent better performance than leading learning-based methods under the same reconstruction error.

Problem

• Improves controllability and accuracy in generative data compression
• Reduces storage costs via latent diffusion and keyframe compression
• Achieves higher compression ratios than rule-based and learning-based methods

Innovation

• Latent diffusion framework combines VAE and diffusion model
• Compresses keyframes into latent space for conditioning
• Generative interpolation reconstructs frames without full storage