From Images to Signals: Are Large Vision Models Useful for Time Series Analysis?

2025-05-29

AI Summary

This work systematically investigates the applicability of large vision models (LVMs) to time-series analysis. To address classification and forecasting—two fundamental tasks—we establish a unified evaluation framework encompassing four LVMs, eight image-based encodings (e.g., Gramian Angular Field, Markov Transition Field), 18 benchmark datasets, and 26 baselines—the first principled empirical study of its kind. Results show that LVMs achieve robust performance in time-series classification, yielding average accuracy gains of 3.2–9.7% over strong baselines. In contrast, they exhibit intrinsic limitations in forecasting: performance degrades significantly with increasing lookback window length, constrained by periodicity bias and insufficient modeling of long-range dependencies—yielding marginal error reduction (<1.1%). The study identifies critical task-specific boundaries between image-based representations and visual architectures, revealing fundamental mismatches in inductive biases. These findings provide key empirical evidence and methodological guidance for designing time-series foundation models.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Transformer-based models have gained increasing attention in time series research, driving interest in Large Language Models (LLMs) and foundation models for time series analysis. As the field moves toward multi-modality, Large Vision Models (LVMs) are emerging as a promising direction. In the past, the effectiveness of Transformer and LLMs in time series has been debated. When it comes to LVMs, a similar question arises: are LVMs truely useful for time series analysis? To address it, we design and conduct the first principled study involving 4 LVMs, 8 imaging methods, 18 datasets and 26 baselines across both high-level (classification) and low-level (forecasting) tasks, with extensive ablation analysis. Our findings indicate LVMs are indeed useful for time series classification but face challenges in forecasting. Although effective, the contemporary best LVM forecasters are limited to specific types of LVMs and imaging methods, exhibit a bias toward forecasting periods, and have limited ability to utilize long look-back windows. We hope our findings could serve as a cornerstone for future research on LVM- and multimodal-based solutions to different time series tasks.

Problem

• Assess LVMs' utility for time series classification and forecasting
• Evaluate LVMs across diverse tasks, datasets, and baseline models
• Identify limitations of LVMs in forecasting and long-range analysis

Innovation

• Utilizing Large Vision Models for time series
• Testing 4 LVMs with 8 imaging methods
• LVMs effective in classification, limited in forecasting