IoU-PD: IoU-Aware Privileged Distillation for Visual Grounding with Multimodal Large Language Models
Authors: Xiuyuan Zhu, Ke Lu, Hao Wu, Siwen Jiao, Zijin Du, Dongming Zhang, Jian Xue
Organizations: University of Chinese Academy of Sciences, Beijing, China · State Key Laboratory of Communication Content Cognition, Beijing, China · Peng Cheng Laboratory, Shenzhen, Guangdong, China · National University of Singapore, Singapore
Visual grounding with multimodal large language models is commonly formulated as autoregressive coordinate generation, where a model outputs bounding-box coordinates as text given an image and a referring-expression prompt. While this interface is simple and compatible with instruction following, it introduces a mismatch between training and evaluation: training optimizes token-level likelihood over coordinate strings, whereas grounding quality is measured by geometric overlap. We propose IoU-PD, an IoU-aware privileged distillation method for coordinate-generating multimodal large language models. IoU-PD uses ground-truth boxes not only as coordinate targets, but also as privileged training-time guidance. During training, the student receives the original image and prompt, while a frozen teacher receives a box-marked image and an augmented prompt that indicates the marked region. The student is trained with a supervised fine-tuning anchor and a privileged distillation loss whose token weights reflect both geometric importance and teacher reliability. At inference time, IoU-PD requires no box overlay, privileged hint, teacher branch, or additional prediction module. Experiments on standard referring-expression grounding benchmarks show consistent region-level improvements over strong coordinate-generating baselines, demonstrating that ground-truth boxes can provide useful privileged guidance beyond serving as coordinate labels.
While Multimodal Large Language Models (MLLMs) excel in cross-modal reasoning, they often struggle to perceive fine-grained details in complex high-resolution images. Recent training-free methods address this through image scaling and localized cropping. However, applying these manipulations indiscriminately introduces computational redundancy for simple queries and can degrade accuracy by truncating essential global context or introducing irrelevant background noise. To this end, we propose LazyMCoT, a dynamic and training-free framework that adaptively allocates visual grounding efforts based on sample difficulty. The framework features an Adaptive Routing mechanism that evaluates predictive uncertainty using first-token statistics from a single forward pass. This efficiently bypasses confident cases while ensuring the recall of difficult samples via conformal calibration. For these challenging cases, a Collaborative Grounding module integrates the inherent cross-modal attention of the model with an external visual expert through a two-stage refinement process. This refinement process generates a precise localized display to recover small or occluded targets. Extensive experiments across diverse benchmarks demonstrate that LazyMCoT rivals training-based approaches by simultaneously improving reasoning accuracy and reducing average inference latency. Our code is availble at https://github.com/TencentBAC/LazyMCoT.
Multi-task visual grounding requires models to jointly understand linguistic semantics and perform accurate visual localization and segmentation. Despite the success of multimodal large language models, effectively adapting them to multiple grounding objectives remains challenging. Existing methods commonly enforce task cooperation through shared representations, while overlooking the intrinsic conflict between task-oriented feature interests. In this paper, we introduce DeCo, an efficient Decouple-to-Couple learning framework that resolves this dilemma through a two-stage paradigm: task-specific representation decoupling followed by complementary prior coupling. Specifically, we first propose Task-aware Semantic Decoupling (TSD) to route shared visual cues into individual features under salient word-level guidance, alleviating representation interference between localization and segmentation. Furthermore, we observe that segmentation naturally provides informative localization priors due to dense supervision. Based on this insight, we introduce Hybrid Prior Coupling (HPC), which integrates sentence-level semantic prior with mask-derived spatial prior for enhanced grounding. Built upon a frozen multimodal encoder, DeCo requires lightweight trainable parameters while achieving strong generalization across multiple grounding objectives. Extensive experiments on RefCOCO/+, G-Ref, ReferIt, Flickr, DIOR-RSVG, SARVG1.0, RRSIS-D, RIS-LAD, and RefDIOR demonstrate that DeCo achieves state-of-the-art performance on both natural and remote sensing benchmarks. The code and models are available at https://github.com/xiaoqiang-lu/DeCo.
Generative Vision-Language Models (VLMs) commonly treat bounding-box coordinates as independent output symbols, leaving numerical order and axis semantics implicit. We identify this representation as an important source of error in visual grounding. Hi-Token encodes each coordinate with axis-specific tokens for the hundreds, tens, and ones digits, which adds coarse-to-fine structure and increases token reuse while retaining the existing VLM architecture. Hi-GAR complements this representation with a geometry-based reward for Group Relative Policy Optimization (GRPO), using box overlap and coordinate accuracy at multiple scales. Controlled comparisons under matched training conditions show that Hi-Token improves localization throughout the evaluated IoU range. Hi-GAR further reduces low-overlap predictions and is used only during training. Experiments on three VLM backbones and the RefCOCO family show consistent gains across models and benchmarks. Hi-R1 achieves higher values than strong specialist baselines on most reported metrics. Analyses of token frequency, digit boundaries, object scale, and IoU distributions explain the effects of coordinate representation and reward training. The results show that structured coordinate generation provides an effective approach to generative visual grounding.