Show, Don't Tell: Evaluating Spatial Cognition in Generative Pixels Rather Than LLM Text

2026-07-23

AI Summary

Current spatial reasoning benchmarks rely on textual or coordinate-based outputs, making it difficult to fairly evaluate the ability of image generation models to express spatial relationships directly through pixels. To address this limitation, this work proposes ProVisE, a novel framework that establishes the first spatial cognition evaluation paradigm specifically designed for image generation models. ProVisE employs protocol-constrained prompting to guide models in producing structured visual responses and introduces a parsing mechanism to map these pixel-level outputs into predictions compatible with conventional evaluation metrics. Additionally, the authors construct SpatialGen-Bench, a multi-granularity diagnostic benchmark that enables unified assessment of both text-based and image-based models. Experimental results reveal that image generation models excel at pixel-level spatial tasks, while text-based vision-language models retain advantages in compositional reasoning, highlighting their complementary strengths.

Abstract

Spatial intelligence is essential for agents to move from static semantic understanding toward interacting with the physical world. Many spatial tasks are grounded in continuous visual scenes, where locations, regions, and paths are more naturally expressed by pointing, marking, or drawing than by reporting precise coordinates or discrete textual symbols. Yet existing spatial reasoning benchmarks usually require coordinates, options, or text, creating an answer-interface mismatch for image-generation models. This makes it difficult to evaluate image-generation models under the same task semantics as text-output VLMs, despite their ability to externalize spatial judgments directly in pixel space. We propose ProVisE (Protocolized Visual Evaluation), a benchmark-agnostic framework that elicits protocol-constrained visual answers from image-generation models and parses them into structured predictions compatible with original metrics. ProVisE also includes an Agentic builder that constructs and validates task-specific protocols for new benchmarks. We further introduce SpatialGen-Bench, a curated diagnostic benchmark of 470 samples across 14 spatial subtasks, four capability levels, and diverse answer forms. We evaluate representative text-output VLMs and image-generation models in a unified setting and validate Agentic protocol construction on six external spatial benchmarks. Results show that image-generation models are competitive when spatial answers can be externalized directly in pixel space, while text-output VLMs retain a clear advantage in compositional spatial reasoning. These findings reveal complementary strengths of pixel-space expression and text-based reasoning and establish a metric-compatible testbed for studying spatial cognition in image-generation models.

Problem

• spatial reasoning
• image-generation models
• visual evaluation
• answer-interface mismatch
• spatial cognition

Innovation

• ProVisE
• spatial cognition
• image-generation models
• visual evaluation
• protocol-constrained output