Test-Time Computing for Referring Multimodal Large Language Models

TL;DR

ControlMLLM++ introduces a test-time adaptation method for multimodal large language models that uses visual prompts and attention steering to enable fine-grained visual reasoning without retraining.

Contribution

It presents a novel framework that injects learnable visual prompts during inference, improving reasoning and interpretability without model fine-tuning.

Findings

Effective across diverse visual prompt types

Strong out-of-domain generalization

Enhanced interpretability of visual reasoning

Abstract

We propose ControlMLLM++, a novel test-time adaptation framework that injects learnable visual prompts into frozen multimodal large language models (MLLMs) to enable fine-grained region-based visual reasoning without any model retraining or fine-tuning. Leveraging the insight that cross-modal attention maps intrinsically encode semantic correspondences between textual tokens and visual regions, ControlMLLM++ optimizes a latent visual token modifier during inference via a task-specific energy function to steer model attention towards user-specified areas. To enhance optimization stability and mitigate language prompt biases, ControlMLLM++ incorporates an improved optimization strategy (Optim++) and a prompt debiasing mechanism (PromptDebias). Supporting diverse visual prompt types including bounding boxes, masks, scribbles, and points, our method demonstrates strong out-of-domain generalization and interpretability. The code is available at https://github.com/mrwu-mac/ControlMLLM.