analiza aiqbench
This research introduces Imaginative Perception Tokens (IPT) as an intermediate representation layer designed specifically to resolve spatial reasoning deficits in vision language models when direct observation is unavailable. By training on tasks like perspective taking and path tracing, IPT enables VLMs to infer unseen viewpoints without generating images at inference time or relying solely on text-based chain-of-thought methods that can degrade performance due to modality mismatches. The findings suggest a more robust architectural approach for handling occluded spaces and integrating partial observations into coherent spatial representations.
Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models
arXiv:2606.03988v3 Announce Type: replace Abstract: Vision language models (VLMs) excel at many tasks but still struggle with spatial reasoning when critical information is not directly observable. Many such problems require imaginative perception: inferring what would be seen from an unseen viewpoint, tracing paths through occluded spaces, or integrating partial observations into a coherent spatial representation. We introduce Imaginative Perception Tokens (IPT), intermediate perceptual representations that externalize what a VLM would perceive under alternative spatial configurations while remaining consistent with the observed input. To study this capability, we formulate three tasks, Perspective Taking (PET), Path Tracing (PT), and Multiview Counting (MVC), and construct datasets of approximately 20K examples with ground truth imaginations, answers, and evaluation benchmarks. Using the unified VLM BAGEL as the backbone, IPT supervision consistently improves spatial reasoning and often outperforms textual chain of thought training, even without generating images at inference time. On MVC, IPT improves accuracy by 3.4% and achieves competitive performance with strong closed-source models on PT. We further find that combining IPT and label-only supervision yields additional gains, whereas textual chain of thought can substantially degrade performance, suggesting a modality mismatch when spatial computation is forced through language. Overall, IPT provides a principled supervision signal for reasoning about unobserved spatial structure, improving generalization while producing interpretable intermediate representations.
Abstract:Vision language models (VLMs) excel at many tasks but still struggle with spatial reasoning when critical information is not directly observable. Many such problems require imaginative perception: inferring what would be seen from an unseen viewpoint, tracing paths through occluded spaces, or integrating partial observations into a coherent spatial representation. We introduce Imaginative Perception Tokens (IPT), intermediate perceptual representations that externalize what a VLM would perceive under alternative spatial configurations while remaining consistent with the observed input. To study this capability, we formulate three tasks, Perspective Taking (PET), Path Tracing (PT), and Multiview Counting (MVC), and construct datasets of approximately 20K examples with ground truth imaginations, answers, and evaluation benchmarks. Using the unified VLM BAGEL as the backbone, IPT supervision consistently improves spatial reasoning and often outperforms textual chain of thought training, even without generating images at inference time. On MVC, IPT improves accuracy by 3.4% and achieves competitive performance with strong closed-source models on PT. We further find that combining IPT and label-only supervision yields additional gains, whereas textual chain of thought can substantially degrade performance, suggesting a modality mismatch when spatial computation is forced through language. Overall, IPT provides a principled supervision signal for reasoning about unobserved spatial structure, improving generalization while producing interpretable intermediate representations.
Artificial Intelligence (cs.AI)
arXiv:2606.03988 [cs.AI]
(or arXiv:2606.03988v3 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2606.03988 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 300 cuvinte
Subiecte conexe
Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models
arXiv · 4 iun. 2026
Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models
arXiv · 3 iun. 2026
Scaling Native Multimodal Pre-Training From Scratch
arXiv · 27 iul. 2026
GeoAnchor: Collaborative Reasoning via Latent Decomposition for 3D Spatial Understanding
arXiv · 29 iul. 2026