VLM Adaptation

VLM: Vision-Language Model

Follow this topic Ask about this topic

Vision-language model adaptation studies how to specialize pretrained models such as CLIP and instruction-tuned multimodal large language models when target data, labels, or compute are limited. Common approaches include prompt tuning, LoRA and lightweight adapters, domain-specific fine-tuning, and test-time methods that use unlabeled examples or prediction memories; newer work also adapts models for video, medical imaging, remote sensing, and robotics. Evaluation often tests few-shot transfer, base-to-novel classes, cross-dataset generalization, robustness to distribution shift, and retention of the model’s original abilities. A central tension is that stronger specialization can cause forgetting, miscalibration, or reliance on language shortcuts, while conservative updates may preserve generality but limit task gains.

Latest papers

546 papers41 papers in the last four weeks, down 13% on the four weeks before. 0.2% of all new papers.

Apr 16, 2026

AIM: Asymmetric Information Masking for Visual Question Answering Continual Learning

In continual visual question answering (VQA), existing Continual Learning (CL) methods are mostly built for symmetric, unimodal architectures. However, modern Vision-Language Models (VLMs) violate this assumption, as their trainable components are inherently asymmetric. This structural mismatch renders VLMs highly prone to catastrophic forgetting when learning from continuous data streams. Specifically, the asymmetry causes standard global regularization to favor the massive language decoder during optimization, leaving the smaller but critical visual projection layers highly vulnerable to interference. Consequently, this localized degradation leads to a severe loss of compositional reasoning capabilities. To address this, we propose Asymmetric Information Masking (AIM), which balances stability and plasticity by applying targeted masks based on modality-specific sensitivity. Experiments on VQA v2 and GQA under continual VQA settings show that AIM achieves state-of-the-art performance in both Average Performance (AP) and Average Forgetting (AF), while better preserving generalization to novel skill-concept compositions.

Peifeng Zhang, Zice Qiu, Donghua Yu +4

Sun Yat-Sen University · National Supercomputing Center in Shenzhen · Tsinghua University

Visual Question Answering Vision-Language Models VLM Adaptation Continual Learning for VLMs Catastrophic Forgetting

Apr 15, 2026

BARD: Bridging AutoRegressive and Diffusion Vision-Language Models Via Highly Efficient Progressive Block Merging and Stage-Wise Distillation

Autoregressive vision-language models (VLMs) deliver strong multimodal capability, but their token-by-token decoding imposes a fundamental inference bottleneck. Diffusion VLMs offer a more parallel decoding paradigm, yet directly converting a pretrained autoregressive VLM into a large-block diffusion VLM (dVLM) often leads to substantial quality degradation. In this work, we present BARD, a simple and effective bridging framework that converts a pretrained autoregressive VLM into a same-architecture, decoding-efficient dVLM. Our approach combines progressive supervised block merging, which gradually enlarges the decoding block size, with stage-wise intra-dVLM distillation from a fixed small-block diffusion anchor to recover performance lost at larger blocks. We further incorporate a mixed noise scheduler to improve robustness and token revision during denoising, and memory-friendly training to enable efficient training on long multimodal sequences. A key empirical finding is that direct autoregressive-to-diffusion distillation is poorly aligned and can even hurt performance, whereas distillation within the diffusion regime is consistently effective. Experimental results show that, with

\leq

4.4M data, BARD-VL transfers strong multimodal capability from Qwen3-VL to a large-block dVLM. Remarkably, BARD-VL establishes a new SOTA among comparable-scale open dVLMs on our evaluation suite at both 4B and 8B scales. At the same time, BARD-VL achieves up to 3

\times

decoding throughput speedup compared to the source model. Code is available at https://github.com/fudan-generative-vision/Bard-VL.

Baoyou Chen, Hanchen Xia, Peng Tu +5

Shanghai Academy of AI for Science · Fudan University · Shanghai Innovation Institute +1

Vision-Language Models Diffusion VLMs VLM Adaptation Diffusion Model Distillation Efficient VLM Inference VLM Distillation+1

Apr 14, 2026

MODIX: A Training-Free Multimodal Information-Driven Positional Index Scaling for Vision-Language Models

Vision-Language Models (VLMs) have achieved remarkable progress in multimodal understanding, yet their positional encoding mechanisms remain suboptimal. Existing approaches uniformly assign positional indices to all tokens, overlooking variations in information density within and across modalities, which leads to inefficient attention allocation where redundant visual regions dominate while informative content is underrepresented. We identify positional granularity as an implicit resource and propose MODIX (Multimodal Information-Driven Positional IndeX Scaling), a training-free framework that dynamically adapts positional strides based on modality-specific contributions. MODIX jointly models intra-modal density via covariance-based entropy and inter-modal interaction via cross-modal alignment to derive unified scores, which rescale positional indices to allocate finer granularity to informative modalities while compressing redundant ones, without requiring any modification to model parameters or architecture. Experiments across diverse architectures and benchmarks demonstrate that MODIX consistently improves multimodal reasoning and adaptively reallocates attention according to task-dependent information distributions, suggesting that positional encoding should be treated as an adaptive resource in Transformers for multimodal sequence modeling.

Ruoxiang Huang, Zhen Yuan

Peking University, China

Vision-Language Models VLM Adaptation Multimodal Reasoning

Apr 14, 2026

Dual-Modality Anchor-Guided Filtering for Test-time Prompt Tuning

Test-Time Prompt Tuning (TPT) adapts vision-language models using augmented views, but its effectiveness is hindered by the challenge of determining which views are beneficial. Standard entropy-based filtering relies on the internal confidence scores of the model, which are often miscalibrated under distribution shift, assigning high confidence to irrelevant crops or background regions while ignoring semantic content. To address this, we propose a dual-modality anchor-guided framework that grounds view selection in semantic evidence. We introduce a text anchor from attribute-rich descriptions, to provide fine-grained class semantics, and an adaptive image anchor that captures evolving test-time statistics. Using these anchors, we filter views based on alignment and confidence, ensuring that only informative views guide adaptation. Moreover, we treat the anchors as auxiliary predictive heads and combine their predictions with the original output in a confidence-weighted ensemble, yielding a stable supervision signal for prompt updates. Extensive experiments on 15 benchmark datasets demonstrate new state-of-the-art performance, highlighting the contribution of anchor-guided supervision as a foundation for robust prompt updates.

Jungwon Choi, Eunwoo Kim

School of Computer Science and Engineering, Chung-Ang University

VLM Robustness VLM Adaptation Test-Time Prompt Tuning

Apr 14, 2026

Unlocking the Potential of Grounding DINO in Videos: Parameter-Efficient Adaptation for Limited-Data Spatial-Temporal Localization

Spatio-temporal video grounding (STVG) aims to localize queried objects within dynamic video segments. Prevailing fully-trained approaches are notoriously data-hungry. However, gathering large-scale STVG data is exceptionally challenging: dense frame-level bounding boxes and complex temporal language alignments are prohibitively expensive to annotate, especially for specialized video domains. Consequently, conventional models suffer from severe overfitting on these inherently limited datasets, while zero-shot foundational models lack the task-specific temporal awareness needed for precise localization. To resolve this small-data challenge, we introduce ST-GD, a data-efficient framework that adapts pre-trained 2D visual-language models (e.g., Grounding DINO) to video tasks. To avoid destroying pre-trained priors on small datasets, ST-GD keeps the base model frozen and strategically injects lightweight adapters (~10M trainable parameters) to instill spatio-temporal awareness, alongside a novel temporal decoder for boundary prediction. This design naturally counters data scarcity. Consequently, ST-GD excels in data-scarce scenarios, achieving highly competitive performance on the limited-scale HC-STVG v1/v2 benchmarks, while maintaining robust generalization on the VidSTG dataset. This validates ST-GD as a powerful paradigm for complex video understanding under strict small-data constraints.

Zanyi Wang, Fan Li, Dengyang Jiang +4

SGIT AI Lab, State Grid Corporation of China · University of HongKong · Zhejiang University of Technology

Temporal Video Grounding VLM Adaptation Parameter-Efficient Fine-Tuning

Apr 13, 2026

Anthropogenic Regional Adaptation in Multimodal Vision-Language Model

While the field of vision-language (VL) has achieved remarkable success in integrating visual and textual information across multiple languages and domains, there is still no dedicated framework for assessing human-centric alignment in vision-language systems. We offer two contributions to address this gap. First, we introduce Anthropogenic Regional Adaptation: a novel paradigm that aims to optimize model relevance to specific regional contexts while ensuring the retention of global generalization capabilities. Second, we present a simple, but effective adaptation method named Geographical-generalization-made-easy (GG-EZ), which utilizes regional data filtering and model merging. Through comprehensive experiments on 3 VL architectures: large vision-language models, text-to-image diffusion models, and vision-language embedding models, and a case study in Southeast Asia (SEA) regional adaptation, we demonstrate the importance of Anthropogenic Regional Adaptation and the effectiveness of GG-EZ, showing 5-15% gains in cultural relevance metrics across SEA while maintaining over 98% of global performance and even occasionally surpassing it. Our findings establish Anthropogenic Regional Alignment as a foundational paradigm towards applicability of multimodal vision-language models in diverse regions and demonstrate a simple-yet-effective baseline method that optimizes regional value alignment while preserving global generalization.

Samuel Cahyawijaya, Peerat Limkonchotiwat, Tack Hwa Wong +45

VLM Adaptation Vision-Language Alignment

Apr 13, 2026

Lost in Adaptation: Layer-Selective Recovery of Temporal Reasoning in Video-Language Models

Multimodal adaptation can weaken temporal reasoning (TR) in video-language models (VLMs) even when the visual evidence needed for an answer is correctly perceived. This weakened competence, however, remains partly recoverable from the model's paired pre-adaptation language checkpoint. We introduce MERIT, a gradient-free framework that selectively restores this competence across the language backbone while protecting temporal perception (TP). It assigns each self-attention layer a VLM-dominant or LLM-dominant interpolation and uses the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) to search the recipe space under a perception-constrained objective. Across three VLM families and five video benchmarks, MERIT improves reasoning-aligned performance in all 15 model--benchmark evaluations while preserving TP across all three Video-MME settings. Recipes selected using 232 diagnostic examples transfer to four unseen benchmarks, with relative gains up to 27.8%, and outperform full-backbone, uniform-attention, and alpha-tuned random-layer merging. Interventional masking shows that MERIT-selected layers are disproportionately important for TR, while frame-level attribution reveals increased reliance on temporally distributed, causally relevant evidence. These results show that adaptation-induced reasoning degradation can be repaired post hoc through selective reuse of pre-adaptation competence, without learning new parameters or adding inference overhead.

Zihang Fu, Haonan Wang, Jian Kang +2

National University of Singapore · MBZUAI

Video-Language Models VLM Adaptation Model Merging

Apr 12, 2026

A Progressive Training Strategy for Embodied Vision-Language Models to Mitigate Spatio-Temporal Hallucinations

Vision-Language Models (VLMs) have made significant strides in static image understanding but continue to face critical hurdles in spatiotemporal reasoning. A major bottleneck is "multi-image reasoning hallucination", where a large performance drop between forward and reverse temporal queries reveals a dependence on superficial shortcuts instead of state-based understanding. To mitigate this, we first develop a new Chain-of-Thought (CoT) dataset that decomposes intricate reasoning into detailed spatiotemporal steps and definitive judgments. Building on this, we present a progressive training framework: it initiates with supervised pre-training on our CoT dataset to instill logical structures, followed by fine-tuning with scalable weakly-labeled data for broader generalization. Our experiments demonstrate that this approach not only improves backbone accuracy but also reduces the forward-backward performance gap from over 70% to only 6.53%. This shows that the method strengthens dynamic reasoning and reduces the inherent temporal biases of current VLMs.

Xiaoda Yang, Shuai Yang, Can Wang +9

The Hong Kong University of Science and Technology · The Chinese University of Hong Kong

VLM Adaptation VLM Hallucination VLM Reasoning Temporal Video Understanding

Apr 10, 2026

Seeing is Believing: Robust Vision-Guided Cross-Modal Prompt Learning under Label Noise

Prompt learning is a parameter-efficient approach for vision-language models, yet its robustness under label noise is less investigated. Visual content contains richer and more reliable semantic information, which remains more robust under label noise. However, the prompt itself is highly susceptible to label noise. Motivated by this intuition, we propose VisPrompt, a lightweight and robust vision-guided prompt learning framework for noisy-label settings. Specifically, we exploit a cross-modal attention mechanism to reversely inject visual semantics into prompt representations. This enables the prompt tokens to selectively aggregate visual information relevant to the current sample, thereby improving robustness by anchoring prompt learning to stable instance-level visual evidence and reducing the influence of noisy supervision. To address the instability caused by using the same way of injecting visual information for all samples, despite differences in the quality of their visual cues, we further introduce a lightweight conditional modulation mechanism to adaptively control the strength of visual information injection, which strikes a more robust balance between text-side semantic priors and image-side instance evidence. The proposed framework effectively suppresses the noise-induced disturbances, reduce instability in prompt updates, and alleviate memorization of mislabeled samples. VisPrompt significantly improves robustness while keeping the pretrained VLM backbone frozen and introducing only a small amount of additional trainable parameters. Extensive experiments under synthetic and real-world label noise demonstrate that VisPrompt generally outperforms existing baselines on seven benchmark datasets and achieves stronger robustness. Our code is publicly available at https://github.com/gezbww/Vis_Prompt.

Zibin Geng, Xuefeng Jiang, Jia Li +6

Institute of Computing Technology, Chinese Academy of Sciences, University of Chinese Academy of Sciences, Beijing, China · Institute of Information Engineering, Chinese Academy of Sciences, University of Chinese Academy of Sciences, Beijing, China · PCALab, VCIP, College of Computer Science, Nankai University, Tianjin, China +1

VLM Robustness Visual Prompt Learning VLM Adaptation Noisy-Label Learning Cross-Modal Learning

Apr 10, 2026

VisionFoundry: Teaching VLMs Visual Perception with Synthetic Images

Vision-language models (VLMs) still struggle with visual perception tasks such as spatial understanding and viewpoint recognition, largely because natural image datasets provide limited supervision for low-level visual skills. Can targeted synthetic supervision address these weaknesses without reference images or manual annotation? To investigate this, we introduce VisionFoundry, an automated pipeline that takes only a task name as input, uses LLMs to synthesize paired questions, answers, and text-to-image (T2I) prompts, generates images with T2I models, and filters samples via multimodal verification. With VisionFoundry, we construct VisionFoundry-10k, a synthetic VQA dataset spanning 10 perception tasks. Finetuning on VisionFoundry-10k consistently improves perception benchmarks across three open-source backbones (e.g., +6.7% on MMVP-pair and +10.5% on CV-Bench-3D for Qwen2.5-VL-3B-Instruct) while preserving broader capabilities and showing positive data scaling. The same synthetic supervision also yields consistent gains under reinforcement learning (RL) across all three backbones, and the framework remains effective under open-source synthesis and self-verification. Our findings demonstrate that automated synthetic supervision offers an effective and scalable path toward systematic VLM training.

Guanyu Zhou, Yida Yin, Wenhao Chai +3

Princeton University · New York University

Visual Question Answering Synthetic Data Augmentation Vision-Language Models Visual Spatial Reasoning VLM Adaptation

Apr 10, 2026

Through Their Eyes: Fixation-aligned Tuning for Personalized User Emulation

Large language model (LLM) agents are increasingly deployed as scalable user simulators for recommender system evaluation. Yet existing simulators perceive recommendations through text or structured metadata rather than the visual interfaces real users browse-a critical gap, since attention over recommendation layouts is both visually driven and highly personalized. We investigate whether aligning a vision-language model's (VLM's) visual attention with user-specific gaze patterns can improve simulation fidelity. Analysis of a real-world eye-tracking dataset collected in a carousel-based recommendation setting reveals that users exhibit stable individual gaze patterns strongly predictive of click behavior. Building on this finding, we propose Fixation-Aligned Tuning for user Emulation (FixATE). Our approach first probes the VLM's internal visual attention via interpretability operators to obtain a slot-level relevance distribution comparable with human fixation, and then learns personalized soft prompts to steer the model's attention toward each user's characteristic fixation pattern. Experiments across three interpretability-based probing operators and two architecturally distinct VLM backbones demonstrate consistent improvements in both attention alignment and click prediction accuracy. These results suggest that making the model "see like the user" is a viable path toward simulators that more faithfully reproduce how users perceive and act in recommendation interfaces.

Lingfeng Huang, Huizhong Guo, Tianjun Wei +2

Singapore University of Technology and Design, Singapore · Zhejiang University, Hangzhou, China · Nanyang Technological University, Singapore

User Simulation VLM Adaptation Personalized Language Model Alignment

Apr 10, 2026

Low-Data Supervised Adaptation Outperforms Prompting for Cloud Segmentation Under Domain Shift

Adapting vision-language models to remote sensing imagery presents a fundamental challenge: both the visual and linguistic distributions of satellite data lie far outside natural image pretraining corpora. Despite this, prompting remains the dominant deployment paradigm, driven by the assumption that domain-specific language can guide frozen model representations toward specialized tasks. We test this assumption directly on a domain where the mismatch is prominent: cloud segmentation for satellite imagery. Using CLIPSeg on the CloudSEN12+ cloud segmentation benchmark, we evaluate 60 prompt variants spanning simple labels, domain terminology, appearance descriptors, and contextual cues, finding that every variant underperforms the zero-shot baseline (0.255 mIoU), with engineered prompts scoring as low as 0.07 mIoU. No amount of linguistic refinement bridges the gap between CLIP's natural image representations and satellite spectral imagery. In contrast, supervised fine-tuning with just 0.1% labeled data (~8 images) surpasses zero-shot performance overall, and 5-10% data recovers ~85% of maximum achievable mIoU. Full fine-tuning consistently outperforms low-rank adaptation by 0.03-0.09 mIoU, with the largest gaps for spectrally ambiguous classes, and at 0.5 to 1% labeled data, fine-tuning temporarily degrades performance on these classes before recovering, a supervision dip that aggregate mIoU can mask. For practitioners adapting vision-language models to specialized imagery, our results deliver a clear message: labeled data is not the expensive alternative to prompting; it is the worthwhile path.

Harshith Kethavath, Weiming Hu

University of Georgia, USA

VLM Adaptation Remote Sensing VLMs Fine-Tuning Image Segmentation Remote Sensing Image Segmentation

Apr 9, 2026

T-Gated Adapter: A Lightweight Temporal Adapter for Vision-Language Medical Segmentation

Medical image segmentation traditionally relies on fully supervised 3D architectures that demand a large amount of dense, voxel-level annotations from clinical experts which is a prohibitively expensive process. Vision Language Models (VLMs) offer a powerful alternative by leveraging broad visual semantic representations learned from billions of images. However, when applied independently to 2D slices of a 3D scan, these models often produce noisy and anatomically implausible segmentations that violate the inherent continuity of anatomical structures. We propose a temporal adapter that addresses this by injecting adjacent-slice context directly into the model's visual token representations. The adapter comprises a temporal transformer attending across a fixed context window at the token level, a spatial context block refining within-slice representations, and an adaptive gate balancing temporal and single-slice features. Training on 30 labeled volumes from the FLARE22 dataset, our method achieves a mean Dice of 0.704 across 13 abdominal organs with a gain of +0.206 over the baseline VLM trained with no temporal context. Zero-shot evaluation on BTCV and AMOS22 datasets yields consistent improvements of +0.210 and +0.230, with the average cross-domain performance drop reducing from 38.0% to 24.9%. Furthermore, in a cross-modality evaluation on AMOS22 MRI with neither model receiving any MRI supervision, our method achieves a mean Dice of 0.366, outperforming a fully supervised 3D baseline (DynUNet, 0.224) trained exclusively on CT, suggesting that CLIP's visual semantic representations generalize more gracefully across imaging modalities than convolutional features.

Pranjal Khadka

Independent Researcher

Medical Image Segmentation 3D Medical Image Segmentation VLM Adaptation Zero-Shot Medical Image Segmentation

Apr 8, 2026

Holistic Optimal Label Selection for Robust Prompt Learning under Partial Labels

Prompt learning has gained significant attention as a parameter-efficient approach for adapting large pre-trained vision-language models to downstream tasks. However, when only partial labels are available, its performance is often limited by label ambiguity and insufficient supervisory information. To address this issue, we propose Holistic Optimal Label Selection (HopS), leveraging the generalization ability of pre-trained feature encoders through two complementary strategies. First, we design a local density-based filter that selects the top frequent labels from the nearest neighbors' candidate sets and uses the softmax scores to identify the most plausible label, capturing structural regularities in the feature space. Second, we introduce a global selection objective based on optimal transport that maps the uniform sampling distribution to the candidate label distributions across a batch. By minimizing the expected transport cost, it can determine the most likely label assignments. These two strategies work together to provide robust label selection from both local and global perspectives. Extensive experiments on eight benchmark datasets show that HopS consistently improves performance under partial supervision and outperforms all baselines. Those results highlight the merit of holistic label selection and offer a practical solution for prompt learning in weakly supervised settings.

Yaqi Zhao, Haoliang Sun, Yating Wang +2

School of Software, Shandong University, Jinan, China

Prompt Learning VLM Adaptation Weakly Supervised Learning

Apr 7, 2026

Lightweight Multimodal Adaptation of Vision Language Models for Species Recognition and Habitat Context Interpretation in Drone Thermal Imagery

This study proposes a lightweight multimodal adaptation framework to bridge the representation gap between RGB-pretrained VLMs and thermal infrared imagery, and demonstrates its practical utility using a real drone-collected dataset. A thermal dataset was developed from drone-collected imagery and was used to fine-tune VLMs through multimodal projector alignment, enabling the transfer of information from RGB-based visual representations to thermal radiometric inputs. Three representative models, including InternVL3-8B-Instruct, Qwen2.5-VL-7B-Instruct, and Qwen3-VL-8B-Instruct, were benchmarked under both closed-set and open-set prompting conditions for species recognition and instance enumeration. Among the tested models, Qwen3-VL-8B-Instruct with open-set prompting achieved the best overall performance, with F1 scores of 0.935 for deer, 0.915 for rhino, and 0.968 for elephant, and within-1 enumeration accuracies of 0.779, 0.982, and 1.000, respectively. In addition, combining thermal imagery with simultaneously collected RGB imagery enabled the model to generate habitat-context information, including land-cover characteristics, key landscape features, and visible human disturbance. Overall, the findings demonstrate that lightweight projector-based adaptation provides an effective and practical route for transferring RGB-pretrained VLMs to thermal drone imagery, expanding their utility from object-level recognition to habitat-context interpretation in ecological monitoring.

Hao Chen, Fang Qiu, Fangchao Dong +3

Geospatial Information Science, The University of Texas at Dallas, Richardson, TX 75080, USA · Department of Landscape Architecture, University of Florida, Gainesville, FL 32611, USA · College of Forestry, Wildlife and Environment, Auburn University, Auburn, AL 36849, USA +1

VLM Adaptation Wildlife Monitoring

Apr 7, 2026

WRF4CIR: Weight-Regularized Fine-Tuning Network for Composed Image Retrieval

Composed Image Retrieval (CIR) task aims to retrieve target images based on reference images and modification texts. Current CIR methods primarily rely on fine-tuning vision-language pre-trained models. However, we find that these approaches commonly suffer from severe overfitting, posing challenges for CIR with limited triplet data. To better understand this issue, we present a systematic study of overfitting in VLP-based CIR, revealing a significant and previously overlooked generalization gap across different models and datasets. Motivated by these findings, we introduce WRF4CIR, a Weight-Regularized Fine-tuning network for CIR. Specifically, during the fine-tuning process, we apply adversarial perturbations to the model weights for regularization, where these perturbations are generated in the opposite direction of gradient descent. Intuitively, WRF4CIR increases the difficulty of fitting the training data, which helps mitigate overfitting in CIR under limited triplet supervision. Extensive experiments on benchmark datasets demonstrate that WRF4CIR significantly narrows the generalization gap and achieves substantial improvements over existing methods.

Yizhuo Xu, Chaojian Yu, Yuanjie Shao +3

Huazhong University of Science and Technology · University of Sydney

Composed Image Retrieval VLM Adaptation

Apr 6, 2026

The Blind Spot of Adaptation: Quantifying and Mitigating Forgetting in Fine-tuned Driving Models

The integration of Vision-Language Models (VLMs) into autonomous driving promises to solve long-tail scenarios, but this paradigm faces the critical and unaddressed challenge of catastrophic forgetting. The very fine-tuning process used to adapt these models to driving-specific data simultaneously erodes their invaluable pre-trained world knowledge, creating a self-defeating paradox that undermines the core reason for their use. This paper provides the first systematic investigation into this phenomenon. We introduce a new large-scale dataset of 180K scenes, which enables the first-ever benchmark specifically designed to quantify catastrophic forgetting in autonomous driving. Our analysis reveals that existing methods suffer from significant knowledge degradation. To address this, we propose the Drive Expert Adapter (DEA), a novel framework that circumvents this trade-off by shifting adaptation from the weight space to the prompt space. DEA dynamically routes inference through different knowledge experts based on scene-specific cues, enhancing driving-task performance without corrupting the model's foundational parameters. Extensive experiments demonstrate that our approach not only achieves state-of-the-art results on driving tasks but also effectively mitigates catastrophic forgetting, preserving the essential generalization capabilities that make VLMs a transformative force for autonomous systems. Data and model are released at FidelityDrivingBench.

Runhao Mao, Hanshi Wang, Yixiang Yang +3

AutoLab, School of Artificial Intelligence, Shanghai Jiao Tong University

VLMs for Autonomous Driving VLM Adaptation Continual Learning for VLMs Autonomous Driving Benchmarks

Apr 5, 2026

Scale-Aware Vision-Language Adaptation for Extreme Far-Distance Video Person Re-identification

Extreme far-distance video person re-identification (ReID) is particularly challenging due to scale compression, resolution degradation, motion blur, and aerial-ground viewpoint mismatch. As camera altitude and subject distance increase, models trained on close-range imagery degrade significantly. In this work, we investigate how large-scale vision-language models can be adapted to operate reliably under these conditions. Starting from a CLIP-based baseline, we upgrade the visual backbone from ViT-B/16 to ViT-L/14 and introduce backbone-aware selective fine-tuning to stabilize adaptation of the larger transformer. To address noisy and low-resolution tracklets, we incorporate a lightweight temporal attention pooling mechanism that suppresses degraded frames and emphasizes informative observations. We retain adapter-based and prompt-conditioned cross-view learning to mitigate aerial-ground domain shifts, and further refine retrieval using improved optimization and k-reciprocal re-ranking. Experiments on the DetReIDX stress-test benchmark show that our approach achieves mAP scores of 46.69 (A2G), 41.23 (G2A), and 22.98 (A2A), corresponding to an overall mAP of 35.73. These results show that large-scale vision-language backbones, when combined with stability-focused adaptation, significantly enhance robustness in extreme far-distance video person ReID.

Ashwat Rajbhandari, Bharatesh Chakravarthi

Arizona State University

Person Re-Identification VLM Adaptation

Apr 5, 2026

Gram-Anchored Prompt Learning for Vision-Language Models via Second-Order Statistics

Parameter-efficient prompt learning has become the de facto standard for adapting Vision-Language Models (VLMs) to downstream tasks. Existing approaches predominantly focus on aligning text prompts with first-order visual features (i.e., spatial feature maps). While effective for fine-grained semantic discrimination, we argue that relying solely on first-order information is insufficient for robust adaptation, as these spatially entangled features are highly susceptible to domain shifts and local noise. In this work, we propose Gram-Anchored Prompt Learning (GAPL) for Vision-Language Models via Second-Order Statistics, a framework that synergizes local semantic alignment with global structural consistency. Methodologically, we introduce an additional second-order statistical stream via Gram matrices that augments the standard first-order spatial interaction. By anchoring prompts to these second-order priors, our approach enables language representations to dynamically adapt to statistical distribution shifts across diverse domains. Extensive experiments indicate the effectiveness of the second-order features, and show compelling performances of GAPL on various benchmarks.

Minglei Chen, Weilong Wang, Jiang Duan +1

All authors are with Southwestern University of Finance and Economics, Chengdu, Sichuan, China.

Prompt Learning for VLMs Prompt Learning VLM Adaptation

Apr 3, 2026

Toward an Artificial General Teacher: Procedural Geometry Data Generation and Visual Grounding with Vision-Language Models

We study visual explanation in geometry education as a Referring Image Segmentation (RIS) problem: given a diagram and a natural language description, the task is to produce a pixel-level mask for the referred geometric element. However, existing RIS models trained on natural image benchmarks such as RefCOCO fail catastrophically on geometric diagrams due to the fundamental domain shift between photographic scenes and abstract, textureless schematics. To address the absence of suitable training data, we present a fully automated procedural data engine that generates over 200,000 synthetic geometry diagrams with pixel-perfect segmentation masks and linguistically diverse referring expressions, requiring zero manual annotation. We further propose domain-specific fine-tuning of vision-language models (VLMs), demonstrating that a fine-tuned Florence-2 achieves 49% IoU and 85% Buffered IoU (BIoU), compared to <1% IoU in zero-shot settings. We introduce Buffered IoU, a geometry-aware evaluation metric that accounts for thin-structure localization, and show that it better reflects true segmentation quality than standard IoU. Our results establish a foundation for building Artificial General Teachers (AGTs) capable of providing visually grounded, step-by-step explanations of geometry problems.

Hai Nguyen-Truong, Alper Balbay, Tunga Bayrak

Freya

Referring Image Segmentation Visual Reasoning Educational Technology VLM Adaptation Synthetic Data Generation

Apr 1, 2026

Prime Once, then Reprogram Locally: An Efficient Alternative to Black-Box Service Model Adaptation

Adapting closed-box service models (i.e., APIs) for target tasks typically relies on reprogramming via Zeroth-Order Optimization (ZOO). However, this standard strategy is known for extensive, costly API calls and often suffers from slow, unstable optimization. Furthermore, we observe that this paradigm faces new challenges with modern APIs (e.g., GPT-4o). These models can be less sensitive to the input perturbations ZOO relies on, thereby hindering performance gains. To address these limitations, we propose an Alternative efficient Reprogramming approach for Service models (AReS). Instead of direct, continuous closed-box optimization, AReS initiates a single-pass interaction with the service API to prime an amenable local pre-trained encoder. This priming stage trains only a lightweight layer on top of the local encoder, making it highly receptive to the subsequent glass-box (white-box) reprogramming stage performed directly on the local model. Consequently, all subsequent adaptation and inference rely solely on this local proxy, eliminating all further API costs. Experiments demonstrate AReS's effectiveness where prior ZOO-based methods struggle: on GPT-4o, AReS achieves a +27.8% gain over the zero-shot baseline, a task where ZOO-based methods provide little to no improvement. Broadly, across ten diverse datasets, AReS outperforms state-of-the-art methods (+2.5% for VLMs, +15.6% for standard VMs) while reducing API calls by over 99.99%. AReS thus provides a robust and practical solution for adapting modern closed-box models.

Yunbei Zhang, Chengyi Cai, Feng Liu +1

Tulane University · University of Melbourne

VLM Adaptation Vision Foundation Model Adaptation

Apr 1, 2026

Lightweight Prompt-Guided CLIP Adaptation for Monocular Depth Estimation

Leveraging the rich semantic features of vision-language models (VLMs) like CLIP for monocular depth estimation tasks is a promising direction, yet often requires extensive fine-tuning or lacks geometric precision. We present a parameter-efficient framework, named MoA-DepthCLIP, that adapts pretrained CLIP representations for monocular depth estimation with minimal supervision. Our method integrates a lightweight Mixture-of-Adapters (MoA) module into the pretrained Vision Transformer (ViT-B/32) backbone combined with selective fine-tuning of the final layers. This design enables spatially-aware adaptation, guided by a global semantic context vector and a hybrid prediction architecture that synergizes depth bin classification with direct regression. To enhance structural accuracy, we employ a composite loss function that enforces geometric constraints. On the NYU Depth V2 benchmark, MoA-DepthCLIP achieves competitive results, significantly outperforming the DepthCLIP baseline by improving the

δ_1

accuracy from 0.390 to 0.745 and reducing the RMSE from 1.176 to 0.520. These results are achieved while requiring substantially few trainable parameters, demonstrating that lightweight, prompt-guided MoA is a highly effective strategy for transferring VLM knowledge to fine-grained monocular depth estimation tasks.

Reyhaneh Ahani Manghotay, Jie Liang

Simon Fraser University, Burnaby, Canada · School of Engineering Science, Simon Fraser University, Burnaby, BC, Canada · Eastern Institute of Technology, Ningbo, China +1

PEFT of VLMs Monocular Depth Estimation VLM Adaptation Adapter Tuning

Apr 1, 2026

LinguDistill: Recovering Linguistic Ability in Vision-Language Models via Selective Cross-Modal Distillation

Turning a pretrained language model (LM) into a vision-language model (VLM) through multimodal fine-tuning often erodes its native language ability, a form of catastrophic forgetting that shows up even on text-only tasks. This loss is hard to undo with further fine-tuning, and existing remedies add adapters or alignment modules that increase architectural complexity and inference cost. We propose LinguDistill, an adapter-free knowledge distillation method that uses the original frozen LM as the teacher during multimodal post-training. To let a text-only teacher supervise vision-conditioned outputs, we introduce layer-wise KV-cache sharing, which exposes the teacher to the student's multimodal representations without changing either architecture. We then apply distillation selectively, on language-heavy data only, so the teacher restores linguistic ability while the student keeps its visual grounding on document and OCR tasks. LinguDistill recovers the language and knowledge performance lost during multimodal fine-tuning, matching the original VLM on average over text-only benchmarks (ARC, HellaSwag) and exceeding it on ScienceQA, while keeping vision-heavy performance close to standard fine-tuning. Since the teacher is dropped after training, the final model adds no parameters and no inference cost. More broadly, our results show that a model's own pre-adaptation backbone is a practical teacher for undoing forgetting, suggesting a simple recipe for keeping language ability intact as models are extended to new modalities.

Patrick Amadeus Irawan, Erland Hilman Fuadi, Shanu Kumar +2

Mohamed bin Zayed University of Artificial Intelligence

VLM Adaptation VLM Distillation Cross-Modal Knowledge Distillation Continual Learning for VLMs

Apr 1, 2026

Matched-Learning-Rate Analysis of Attention Drift and Transfer Retention in Fine-Tuned CLIP

CLIP adaptation can improve in-domain accuracy while degrading out-of-domain transfer, but comparisons between Full Fine-Tuning (Full FT) and LoRA are often confounded by different learning-rate conventions. We study how adaptation method and optimization scale jointly shape attention drift and transfer retention in CLIP using a controlled matched-learning-rate comparison of Full FT and LoRA. The completed matrix contains 80 runs on CLIP ViT-B/32 across EuroSAT and Oxford-IIIT Pets, spanning four shared learning rates (

10^{-6}

,

5{\times}10^{-6}

,

10^{-5}

,

5{\times}10^{-5}

) and five seeds, and evaluates attention-drift metrics, best validation accuracy, and adapter-aware CIFAR-100 zero-shot accuracy. Learning rate strongly modulates structural change: on EuroSAT, Full FT moves from mild entropy broadening at

10^{-6}

to marked contraction at

5{\times}10^{-5}

, whereas LoRA remains entropy-positive across the full matched grid. At matched learning rates, LoRA preserves substantially more zero-shot transfer than Full FT, averaging

45.13\%

versus

11.28\%

CIFAR-100 accuracy on EuroSAT and

58.01\%

versus

8.54\%

on Pets. Oxford-IIIT Pets also reveals a regime effect: low-learning-rate LoRA underfits in-domain, so method-only averages can obscure when LoRA becomes competitive. Supporting rollout, patch-to-patch, and CKA analyses are directionally consistent with the controlled matrix. Overall, matched-learning-rate evaluation materially changes the interpretation of Full FT versus LoRA, and attention drift is most useful as a descriptive diagnostic of representation preservation rather than a causal explanation of transfer behavior.

Ruize Xia

Independent Researcher

VLM Adaptation Low-Rank Adaptation Cross-Domain Transfer Learning

Apr 1, 2026

CoLA: Cross-Modal Low-rank Adaptation for Multimodal Downstream Tasks

Foundation models have revolutionized AI, but adapting them efficiently for multimodal tasks, particularly in dual-stream architectures composed of unimodal encoders, such as DINO and BERT, remains a significant challenge. ParameterEfficient Fine-Tuning (PEFT) methods like LowRank Adaptation (LoRA) enable lightweight adaptation, yet they operate in isolation within each modality, limiting their ability in capturing cross-modal interactions. In this paper, we take a step in bridging this gap with Cross-Modal LowRank Adaptation (CoLA), a novel PEFT framework that extends LoRA by introducing a dedicated inter-modal adaptation pathway alongside the standard intra-modal one. This dual-path design enables CoLA to adapt unimodal foundation models to multimodal tasks effectively, without interference between modality-specific and crossmodal learning. We evaluate CoLA across a range of vision-language (RefCOCO, RefCOCO+, RefCOCOg) and audio-visual (AVE, AVS) benchmarks, where it consistently outperforms LORA, achieving a relative gain of around 3% and 2%, respectively, while maintaining parameter efficiency. Notably, CoLA enables the first multitask PEFT framework for visual grounding, bridging a key gap in efficient multimodal adaptation. Code is available at https://github.com/peterwisu/CoLA

Wish Suharitdamrong, Tony Alex, Muhammad Awais +1

Surrey Institute for People-Centred AI, University of Surrey, Guildford, GU2 7XH, UK · Centre for Vision, Speech and Signal Processing (CVSSP), University of Surrey, UK

Cross-Modal Learning VLM Adaptation Low-Rank Adaptation Multimodal Foundation Models Parameter-Efficient Fine-Tuning

Mar 31, 2026

The Geometry of Compromise: Unlocking Generative Capabilities via Controllable Modality Alignment

Vision-Language Models (VLMs) such as CLIP learn a shared embedding space for images and text, yet their representations remain geometrically separated, a phenomenon known as the modality gap. This gap limits tasks requiring cross-modal interchangeability, such as captioning and joint clustering. Existing post-processing approaches can partially improve cross-modal compatibility; however, we show through geometric analysis that they primarily reduce the global centroid offset while leaving the underlying distributional mismatch intact. We decompose the modality gap into a Centroid Gap and a Distribution Gap, and demonstrate that the Distribution Gap is the true predictor of cross-modal task quality (

R^2 = 0.986

), whereas the commonly used Raw Gap is misleading (

R^2 = 0.691

). Motivated by this observation, we propose TPC-CMA (Three-Phase Curriculum for Cross-Modal Alignment), a fine-tuning framework that explicitly reduces both components. The proposed CMA jointly mitigates centroid offsets and reshapes the distributional structure, while a three-phase curriculum with gradient-aware scheduling progressively introduces alignment during training to enable stable optimization. Experiments demonstrate that our method significantly improves cross-modal alignment. With

α_{\text{target}}{=}0.05

, the modality gap is reduced by 66.6% with only 4.84% accuracy drop. Under stronger alignment (

α_{\text{target}}{=}0.5

), the gap is reduced by 82.3%, clustering ARI improves from 0.318 to 0.516, and captioning CIDEr increases by 57.1% over the original model. Our code and pre-trained models will be made publicly available upon acceptance.

Hongyuan Liu, Qinli Yang, Wen Li +6

University of Electronic Science and Technology of China · University of Bristol

Cross-Modal Alignment Modality Gap in VLMs VLM Adaptation

Mar 31, 2026

GazeCLIP: Gaze-Guided CLIP with Adaptive-Enhanced Fine-Grained Language Prompt for Deepfake Attribution and Detection

Current deepfake attribution or deepfake detection works tend to exhibit poor generalization to novel generative methods due to the limited exploration in visual modalities alone. They tend to assess the attribution or detection performance of models on unseen advanced generators, coarsely, and fail to consider the synergy of the two tasks. To this end, we propose a novel gaze-guided CLIP with adaptive-enhanced fine-grained language prompts for fine-grained deepfake attribution and detection (DFAD). Specifically, we conduct a novel and fine-grained benchmark to evaluate the DFAD performance of networks on novel generators like diffusion and flow models. Additionally, we introduce a gaze-aware model based on CLIP, which is devised to enhance the generalization to unseen face forgery attacks. Built upon the novel observation that there are significant distribution differences between pristine and forged gaze vectors, and the preservation of the target gaze in facial images generated by GAN and diffusion varies significantly, we design a visual perception encoder to employ the inherent gaze differences to mine global forgery embeddings across appearance and gaze domains. We propose a gaze-aware image encoder (GIE) that fuses forgery gaze prompts extracted via a gaze encoder with common forged image embeddings to capture general attribution patterns, allowing features to be transformed into a more stable and common DFAD feature space. We build a language refinement encoder (LRE) to generate dynamically enhanced language embeddings via an adaptive-enhanced word selector for precise vision-language matching. Extensive experiments on our benchmark show that our model outperforms the state-of-the-art by 6.56% ACC and 5.32% AUC in average performance under the attribution and detection settings, respectively. Codes are available at https://github.com/Jenine-321/GazeCLIP.

Yaning Zhang, Linlin Shen, Zitong Yu +2

Faculty of Computer Science and Technology, Qilu University of Technology (Shandong Academy of Sciences), Jinan, 250014, China · Computer Vision Institute, College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, 518060, China · National Engineering Laboratory for Big Data System Computing Technology, Shenzhen University +5

Deepfake Detection Digital Forensics VLM Adaptation Vision-Language Pretraining

Mar 31, 2026

Omni-NegCLIP: Enhancing CLIP with Front-Layer Contrastive Fine-Tuning for Comprehensive Negation Understanding

Vision-Language Models (VLMs) have demonstrated strong capabilities across a wide range of multimodal tasks. However, recent studies have shown that VLMs, such as CLIP, perform poorly in understanding negation expressions, which are common in natural language. In this work, we propose Omni-NegCLIP, a fine-tuned CLIP model that improves CLIP's understanding of two types of negation, namely presence-based negation and absence-based negation, which correspond to negated expressions of objects that are actually present in an image and those that may plausibly exist in an image but are in fact absent, respectively, by modifying CLIP's original InfoNCE contrastive loss. Specifically, we design a presence-based contrastive objective that pulls image embeddings closer to their original caption embeddings while pushing them away from the corresponding presence-based negated caption embeddings, and an absence-based contrastive objective that aligns image embeddings with both original and absence-based negated caption embeddings while maintaining a semantic distinction between the two text embeddings. Based on our observation that the front transformer layers of CLIP text encoder have stronger learning ability for negated text than the later layers, we fine-tune the front transformer layers of the CLIP text encoder at each training step using the combined contrastive objective. Experimental results show that, compared with pretrained CLIP, Omni-NegCLIP improves performance on presence-based negation and absence-based negation tasks by up to 52.65% and 12.50%, respectively, without sacrificing general capability in image-text retrieval and even improving it by up to 19.62%. Compared with prior works, Omni-NegCLIP demonstrates a more comprehensive ability to understand multiple types of negation tasks.

Jingqi Xu

University of Southern California, Los Angeles, CA, USA

VLM Adaptation VLM Reasoning Contrastive Learning

Mar 30, 2026

Domain-Invariant Prompt Learning for Vision-Language Models

Large pre-trained vision-language models like CLIP have transformed computer vision by aligning images and text in a shared feature space, enabling robust zero-shot transfer via prompting. Soft-prompting, such as Context Optimization (CoOp), effectively adapts these models for downstream recognition tasks by learning a set of context vectors. However, CoOp lacks explicit mechanisms for handling domain shifts across unseen distributions. To address this, we propose Domain-invariant Context Optimization (DiCoOp), an extension of CoOp optimized for domain generalization. By employing an adversarial training approach, DiCoOp forces the model to learn domain-invariant prompts while preserving discriminative power for classification. Experimental results show that DiCoOp consistently surpasses CoOp in domain generalization tasks across diverse visual domains.

Arsham Gholamzadeh Khoee, Yinan Yu, Robert Feldt

Department of Computer Science and Engineering, Chalmers University of Technology, Gothenburg, Sweden

VLM Adaptation Visual Prompt Learning Domain Generalization

Mar 30, 2026

Hydra: Unifying Document Retrieval and Generation in a Single Vision-Language Model

Visual document understanding typically requires separate retrieval and generation models, doubling memory and system complexity. We present Hydra, a dual-head approach that provides both ColBERT-style late-interaction retrieval and autoregressive generation from a single vision-language model. A single LoRA adapter, trained only for retrieval, is toggled at inference: enabling it produces multi-vector embeddings; disabling it recovers the base model's generation quality, with 426 of 426 language-model weight tensors byte-for-byte identical to a freshly-loaded Qwen3.5-4B. We identify two failure modes that can silently break generation in retrieval-fine-tuned VLMs (attention-mode restoration and lm_head preservation) plus an efficiency requirement (KV-cache-aware decoding); Hydra sidesteps the first two structurally and addresses the third in the decode loop. We release two scales, Hydra-4B and Hydra-0.8B, sharing LoRA hyperparameters (r=32, alpha=32) and optimisation recipe; data mix and projection dim differ across scales. The single-model design cuts peak GPU memory from 28.85 GB to 10.77 GB at 4B (62.7% reduction) and from 5.79 GB to 2.37 GB at 0.8B (59.1%) relative to a co-resident two-model deployment. A controlled ablation finds GritLM-style joint training matches Hydra's retrieval-only training on the evaluated modes while its LoRA-on generation mode collapses. A proof-of-concept on Qwen2.5-Omni-3B preserves generation equivalence on a non-Qwen3.5 backbone and transfers image retrieval within 2-8 pp of Hydra-4B, with zero-shot audio retrieval emerging through the frozen Whisper encoder.

Athos Georgiou

Independent Researcher

Vision-Language Models Visual Document Retrieval VLM Adaptation Vision-Language Retrieval Efficient VLM Inference Multi-Vector Retrieval