Referring Multiple Regions with Large Multimodal Models via Contextual Latent Steering

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 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.

We aim at re-evaluating the very recent, popular Large Multimodal Models (LMMs such as Qwen3-VL) over visual referring, which means in this case, users can interact with LMMs with geometric prompts, such as points or boxes. We start with an observation that models like Qwen3-VL not know where to look (by relative attention) when handling multiple region cases. Our method is entirely training-free, by extracting and making use of corrective, contrastive signals from LMMs generated latents.