Referring Multiple Regions with Large Multimodal Models via Contextual Latent Steering

TL;DR

This paper introduces CSteer, a training-free method that guides large multimodal models to accurately refer to multiple regions in images by pre-computing contextual vectors and editing representations during inference.

Contribution

CSteer enables general LMMs to perform multi-region visual referring without fine-tuning, setting new state-of-the-art results in a training-free manner.

Findings

CSteer outperforms tailored referring LMMs on multiple datasets.

The approach is training-free and does not require architectural modifications.

Experimental results demonstrate improved accuracy in multi-region referring tasks.

Abstract

Large Multimodal Models (LMMs) have recently demonstrated their proficiency in holistic visual comprehension. However, most of them struggle to tackle region-level perception guided by visual prompts, especially for cases where multiple regions are referred simultaneously, or scenarios where global contexts are necessary for precise visual referring. We introduce Contextual Latent Steering (CSteer), a training-free approach for guiding general LMMs to refer multiple regions contextually, without expensive fine-tuning or architectural modifications. CSteer starts with pre-computing contextual vectors that implicitly represent visual referring behaviors, such as differentiation among regions and attention to global contexts, followed by representation editing during inference time. Experimental results on multiple datasets indicate that general LMMs with CSteer outperform tailored referring LMMs in most cases, suggesting a promising solution in training-free, and setting new state-of-the-art for this field. Code is available at https://github.com/xing0047/csteer.git.