Graph-of-Mark: Promote Spatial Reasoning in Multimodal Language Models with Graph-Based Visual Prompting

TL;DR

Graph-of-Mark (GoM) introduces scene graph overlays as visual prompts to significantly enhance the spatial reasoning and zero-shot performance of multimodal language models in visual question answering and localization tasks.

Contribution

GoM is the first pixel-level visual prompting method that incorporates scene graphs, enabling better spatial reasoning in multimodal language models.

Findings

Consistently improves zero-shot spatial reasoning in MLMs.

Enhances accuracy in visual question answering and localization tasks.

Effective across multiple datasets and models.

Abstract

Recent advances in training-free visual prompting, such as Set-of-Mark, have emerged as a promising direction for enhancing the grounding capabilities of multimodal language models (MLMs). These techniques operate by partitioning the input image into object regions and annotating them with marks, predominantly boxes with numeric identifiers, before feeding the augmented image to the MLM. However, these approaches treat marked objects as isolated entities, failing to capture the relationships between them. On these premises, we propose Graph-of-Mark (GoM), the first pixel-level visual prompting technique that overlays scene graphs onto the input image for spatial reasoning tasks. We evaluate GoM across 3 open-source MLMs and 4 different datasets, conducting extensive ablations on drawn components and investigating the impact of auxiliary graph descriptions in the text prompt. Our results demonstrate that GoM consistently improves the zero-shot capability of MLMs in interpreting object positions and relative directions, improving base accuracy in visual question answering and localization up to 11 percentage points.