ProLaViT: Learning Progressive Latent Visual Thoughts in Structured Latent Space
2026-07-03 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionComputation and Language
AI summaryⓘ
The authors present ProLaViT, a new method that helps multimodal large language models think through complex visual problems step-by-step in their own hidden visual representations. Instead of relying on outside tools, the model teaches itself by revisiting its own visual encodings. They created special reasoning techniques to handle spatial and logical tasks, including a way to check reasoning by considering alternate possibilities. Their approach also includes a way to keep these visual thoughts diverse and meaningful. Experiments show that ProLaViT is more accurate and efficient than previous methods on visual reasoning tests.
Multimodal Large Language ModelsVisual ReasoningLatent SpaceSelf-DistillationVisual EncoderCausal Chain ReasoningDialectical ReasoningCounterfactual ThinkingDistance-Weighted Diversity LossFeature Degeneration
Authors
Peiming Li, Yifan Wang, Xiaotian Zhang, Zhiyuan Hu, Shiyu Li, Zheng Wei, Yang Tang
Abstract
Multimodal Large Language Models (MLLMs) have achieved remarkable progress but still struggle with complex visual reasoning tasks requiring multi-step perception and logical deduction. While explicit visual generation incurs prohibitive computational costs, existing latent approaches often rely on external experts or lack rigorous cognitive logic. In this paper, we introduce ProLaViT (Progressive Latent Visual Thought), a framework empowering MLLMs to perform structured visual derivation in the continuous latent space. Unlike works dependent on heterogeneous external models, ProLaViT leverages an endogenous self-distillation mechanism, utilizing the model's own visual encoder to supervise latent thoughts. To facilitate this, we construct a scalable programmatic synthesis pipeline enabling the model to internalize algorithmic precision without inference time tools. We design two reasoning paradigms: (1) Coarse-to-Fine Causal Chain for spatial tasks, guiding attention from global context to local targets. (2) Dialectical Reasoning Chain for logical tasks, incorporating counter-factual thinking for verification. Furthermore, we propose a Distance-Weighted Diversity Loss to impose topology-aware constraints, preventing feature degeneration by enforcing semantic distinctiveness. Extensive experiments demonstrate that ProLaViT outperforms baselines on vision-centric benchmarks, achieving superior accuracy and interpretability with high efficiency.