Imagination Helps Visual Reasoning, But Not Yet in Latent Space

TL;DR

This paper investigates the effectiveness of latent visual reasoning in multimodal models, revealing limited causal influence of latent tokens and proposing a simpler explicit imagination method that outperforms complex latent-space approaches.

Contribution

The study uncovers key disconnections in latent reasoning and introduces CapImagine, a straightforward method that explicitly teaches models to imagine visually using text, improving performance.

Findings

Latent tokens have limited causal impact on final answers.

Perturbations on input and latent tokens minimally affect outcomes.

CapImagine outperforms complex latent-space baselines on vision benchmarks.

Abstract

Latent visual reasoning aims to mimic human's imagination process by meditating through hidden states of Multimodal Large Language Models. While recognized as a promising paradigm for visual reasoning, the underlying mechanisms driving its effectiveness remain unclear. Motivated to demystify the true source of its efficacy, we investigate the validity of latent reasoning using Causal Mediation Analysis. We model the process as a causal chain: the input as the treatment, the latent tokens as the mediator, and the final answer as the outcome. Our findings uncover two critical disconnections: (a) Input-Latent Disconnect: dramatic perturbations on the input result in negligible changes to the latent tokens, suggesting that latent tokens do not effectively attend to the input sequence. (b) Latent-Answer Disconnect: perturbations on the latent tokens yield minimal impact on the final answer, indicating the limited causal effect latent tokens imposing on the outcome. Furthermore, extensive probing analysis reveals that latent tokens encode limited visual information and exhibit high similarity. Consequently, we challenge the necessity of latent reasoning and propose a straightforward alternative named CapImagine, which teaches the model to explicitly imagine using text. Experiments on vision-centric benchmarks show that CapImagine significantly outperforms complex latent-space baselines, highlighting the superior potential of visual reasoning through explicit imagination.