VLM Distillation

VLM: Vision-Language Model

Ask about this topic

Overview

Vision-language model distillation studies how to transfer image-and-text understanding from a teacher model to a smaller or more efficient student, often for tasks such as visual question answering, retrieval, video reasoning, and robotics. Methods align teacher and student features, logits, attention, or relationships between visual and text representations; others train on teacher-generated rationales or student-generated responses, sometimes using multiple teachers or privileged visual cues. Work also targets practical constraints such as fewer visual tokens, quantization, and deployment on devices with limited memory or compute. A central challenge is deciding which teacher signals are reliable and grounded in evidence the student can access, since copying a teacher too closely can preserve its errors or encourage shortcuts rather than visual reasoning.

Latest papers

CardsList

126 papers29 papers in the last four weeks, up 142% on the four weeks before. 0.2% of all new papers.

Jun 4, 2026

ViCuR: Visual Cues as Recoverable Privilege for Multimodal On-Policy Distillation

On-policy distillation (OPD) improves reasoning by training a student on trajectories sampled from its own policy under supervision from a teacher. In multimodal reasoning, a common extension is to use a privileged teacher that observes training-time-only signals such as reference answers or rationales. However, such answer-side privilege creates a train-test mismatch: the teacher's supervision may depend on signals unavailable to the student, encouraging shortcut imitation rather than visually grounded reasoning. We propose ViCuR, a visually grounded privileged-teacher distillation framework that replaces answer-side privilege with visual cues (query-related evidence in the input). Because these cues are derived from the same visual input available at inference, their evidence is recoverable by the student. To support this, ViCuR introduces a lightweight cue recovery module that uses dedicated sink-token cross-attention during prefill to aggregate task-relevant visual evidence into an internal representation, without changing the inference interface or requiring auxiliary cue-generation losses. Across seven benchmarks with Qwen3-VL-2B and 8B students, ViCuR consistently improves over answer-based on-policy self-distillation by +1.19 and +1.24 on overall average performance. It also extends naturally to stronger-teacher OPD, surpassing OPD baselines by +0.64 and +1.08, with consistent out-of-domain gains at the 8B scale. These results show that, in multimodal on-policy distillation, the design of teacher privilege is as important as teacher strength.

Kanghui Tian, Siyuan Liu, Ziang Yan +3

1Shanghai AI Laboratory · 3Nanjing University · 2Fudan University

VLM DistillationMultimodal ReasoningOn-Policy DistillationLearning Using Privileged InformationTeacher-Student LearningKnowledge Distillation
AskPDF
Jun 3, 2026

Geometry-Aware Distillation for Prompt Tuning Biomedical Vision-Language Models

Current prompt-based and adapter-based tuning of vision-language models (VLMs) is attractive for medical imaging, where clinical data sensitivity favors frozen backbones and annotations are limited. However, these methods typically optimize only the ground-truth class, treating all other classes as equally incorrect, ignoring clinically meaningful class relations and yielding unstable decision boundaries in limited-supervision settings. We propose Omni-Geometry Knowledge Distillation (OGKD), a new framework that injects class-relation structure into the teacher to produce directional targets that preserve the ground truth while respecting inter-class geometry. Using these targets, we develop two distillation losses: Global Geometry-Aware Distillation (GAD) operates on the global image token, and Label-Guided Geometry Distillation (LGD) applies the same geometry to attentive patch tokens to improve fine-grained alignment. Across comprehensive experiments and analyses on 11 widely-used medical datasets for base-to-novel and few-shot evaluations, our OGKD achieves substantially better performance, consistently improving accuracy by an average absolute gain of 1.7%-2.8% over all prior state-of-the-art VLM adaptation counterparts. It also robustly generalizes to unseen classes and yields more reliable predictions than other approaches. Our code is available at https://github.com/tientrandinh/OGKD.

Tran Dinh Tien, Zhiqiang Shen

Department of Machine Learning Mohamed bin Zayed University of Artificial Intelligence

VLM AdaptationVLM DistillationMedical VLMsRelational Knowledge DistillationKnowledge DistillationVisual Prompt Learning
AskPDF
May 30, 2026

Decomposed On-Policy Distillation for Vision-Language Reasoning: Steering Gradients for Visual Grounding

While on-policy distillation offers dense supervision for training small reasoning models, its optimization dynamics in the multimodal domain remain under-explored. In this work, we challenge the standard monolithic view of Vision-Language Model (VLM) distillation by mathematically decomposing the loss into two distinct components: the language prior and visual grounding. Our analysis uncovers that gradient vectors for these components are nearly orthogonal, indicating that the objective of aligning with the teacher's language distribution is geometrically independent from the objective of matching its visual perception. Consequently, standard optimization passively follows a suboptimal compromise trajectory that implicitly balances the two objectives. Hypothesizing that visual grounding constitutes the primary bottleneck for vision-language reasoning, we introduce Visual Gradient Steering (VGS), a method that dynamically reorients the update vector to prioritize the visual subspace. Experimental results on multiple distillation settings and complex multimodal benchmarks demonstrate that VGS significantly outperforms the standard monolithic formulation of on-policy distillation, achieving superior grounding with minimal training overhead.

Hee Suk Yoon, Eunseop Yoon, Jaehyun Jang +6

Korea Advanced Institute of Science and Technology (KAIST) · University of Illinois Urbana-Champaign (UIUC) · Microsoft Research Asia (MSRA)

Cross-Modal Knowledge DistillationVLM DistillationVision-Language GroundingVLM ReasoningOn-Policy Distillation
AskPDF
May 22, 2026

Multimodal Distribution Matching for Vision-Language Dataset Distillation

Dataset distillation compresses large training sets into compact synthetic datasets while preserving downstream performance. As modern systems increasingly operate on paired vision-language inputs, multimodal distillation must preserve representation quality and cross-modal alignment under tight compute and memory budgets, yet prior methods often require heavy computes and overlook their correlations. To address this, we present Multimodal Distribution Matching (MDM), a geometry-aware framework for efficient and generalizable multimodal distillation. Specifically, MDM integrates complementary components at the data, model, and loss levels. At the data level, it initializes synthetic image-text pairs by sampling from clusters in the joint embedding space. At the model level, it forms a mixed teacher by interpolating independently fine-tuned models in weight space according to their angular deviation from the pretrained anchor. At the loss level, it matches joint distributions on the unit hypersphere using a geometry-aware matching objective that exploits the joint features in the cross-modal agreement and discrepancy directions along with symmetric contrastive learning. Across image-text retrieval benchmarks with cross-architecture evaluation, MDM yields compact synthetic sets that preserve multimodal semantics, substantially reduce distillation cost, and remain robust across architectures.

Jongoh Jeong, Hoyong Kwon, Minseok Kim +1

Visual Intelligence Lab., KAIST

Cross-Modal LearningCross-Modal AlignmentVLM DistillationDistribution MatchingDataset Distillation
AskPDF
May 21, 2026

Visual-Advantage On-Policy Distillation for Vision-Language Models

On-policy knowledge distillation has proven effective for language models, yet its application to vision-language models (VLMs) remains underexplored. We observe that standard on-policy distillation can improve a student's output quality while failing to strengthen its reliance on visual input: on vision-critical tokens, the student's predictions remain largely unchanged whether or not fine-grained visual detail is present, even though the teacher's predictions depend heavily on it.To make this difference observable, we introduce visual advantage (VA), the token-level log-probability difference when the teacher scores a student-generated rollout with versus without access to fine-grained visual detail. VA is concentrated in a small minority of tokens, and these high-VA tokens are the ones that actually carry the visual supervision signal. This motivates a distillation objective that treats them differently from language scaffolding, so their contribution is not diluted by the abundant surrounding language tokens.We propose Visual-Advantage On-Policy Distillation (VA-OPD), which uses VA at two granularities: rollout-level reweighting by trajectory-averaged VA, and token-level KL averaged within high-VA and low-VA groups separately. We train on two math datasets (Geometry3K and ViRL39K) and evaluate on eight benchmarks covering both mathematical reasoning and visual understanding, across three teacher sizes (4B, 8B, and 32B) on the Qwen3-VL family. VA-OPD improves over standard on-policy distillation on every benchmark, with the gain growing monotonically along both the teacher-size and data-scale axes, suggesting that these factors compound consistently.

Ruiqi Liu, Xiaolei Lv, Gengsheng Li +8

Institute of Automation, CAS · School of Advanced Interdisciplinary Sciences, UCAS · 3Hello +1

Vision-Language ModelsVLM DistillationVLM ReasoningVisually Grounded ReasoningOn-Policy DistillationTeacher-Student Learning+1
AskPDF
May 20, 2026

Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Driving

Current end-to-end autonomous driving models are fundamentally constrained by the behavioral cloning ceiling of imitation learning. While reinforcement learning offers a path to smarter autonomy, it demands two missing pieces of infrastructure: (1) a cognitive foundation that understands traffic semantics and driving intent, and (2) a foresighted physical environment that can anticipate the consequences of candidate actions. To this end, we propose CoPhy, a CognitivePhysical reinforcement learning framework for autonomous driving. To distill to think, we distill VLM knowledge into the BEV encoder and then discard the VLM entirely, retaining cognitive ability at zero inference cost while releasing the cognitive channel as a pluggable interface for optional human language commands. To foresee to act, we build an auto-regressive BEV world model that explicitly predicts future semantic maps conditioned on candidate actions, serving as an interpretable physical sandbox from which safety metrics are directly derived. Built upon this dual infrastructure, we optimize the driving policy via GRPO with a novel dual-reward mechanism: a physical reward derived from BEV rollouts enforces hard safety constraints, while a cognitive reward from a language-aligned scorer ensures intent compliance. Extensive experiments demonstrate that CoPhy not only achieves state-of-the-art results on NAVSIM v1 and v2 benchmarks, but also enables safer driving via cognitively informed scene compliance and flexible intent control through user-defined language instructions.

Yang Wu, Qiang Meng, Zhaojiang Liu +3

SJTU · FDU · NJU

End-to-End Autonomous DrivingNavSimWorld ModelsVLM DistillationWorld Models for Autonomous DrivingRL for Autonomous Driving
AskPDF
May 19, 2026

Preferences Order, Ratings Anchor: From Fused Expert Aesthetic Ground Truth to Self-Distillation

Pairwise preferences and pointwise ratings are the two dominant annotation protocols in image aesthetic assessment (IAA), yet existing benchmarks adopt only one, leaving their complementarity unmeasured under controlled conditions. We introduce PPaint, a matched dual-protocol benchmark in which 15 domain experts, 5 per category, annotate 150 Chinese paintings under both protocols across five aesthetic dimensions, collecting 45,900 pairwise expert judgments through a locally dense preference design alongside the matched ratings. The matched design reveals complementary strengths: preferences yield more consistent ordinal rankings, while ratings anchor the absolute score scale. Fusing both signals via two independent preference-to-score methods yields a fused expert ground truth on which the two constructions converge to nearly identical scores. The same preference-to-score principle extends to label-free VLM training. PSDistill converts VLM pairwise judgments into calibrated pseudo-scores via an Elo reference pool, and trains the same VLM with confidence-weighted ranking optimization to produce a single-pass aesthetic scorer. Trained on a single painting category, the distilled Qwen3-VL-8B improves mean SRCC from 0.504 to 0.709 across all three categories, outperforming all open-source baselines including the dedicated aesthetic model ArtiMuse and matching closed-source Gemini-3.1-Pro within 0.04 SRCC at single-pass inference cost, with cross-domain transfer further validated on APDDv2. We will release the full PPaint dataset and training code.

Yuanpei Zhao, Jie Lin, Chao Zhang +5

1Sichuan University · 2NetEase Fuxi AI Lab · †Work done during an internship at NetEase Fuxi AI Lab. +1

Pairwise Preference LearningPersonalized Image Aesthetics AssessmentPairwise ComparisonPairwise Preference EvaluationVLM DistillationPairwise Ranking+3
AskPDF
May 16, 2026

HEED: Density-Weighted Residual Alignment for Hybrid Vision-Language Model Distillation

Distilling vision-language models into faster hybrid architectures, such as 3:1 Mamba-2/attention mixes, is now standard practice for making inference efficient. Aggregate benchmarks suggest that this works but they hide selective failures. When we distill Qwen3-VL-8B-Instruct into a 3:1 Mamba-2/attention hybrid, student model stays within 2 points of the teacher across visual reasoning benchmarks like MMStar, MMBench, and MMMU-Pro, while dropping 13 points on optical-character-recognition and document tasks. The student can still understand the scene but loses the fine-grained text needed to answer. We localize much of the failure to a specific kind of position. In a high-resolution image, most patches are sky, wall, or smooth texture, while a small fraction carries text, edges, object boundaries, or other local details. In a token-level diagnostic, the top 10% highest-density patches have 3.6×\times larger residual drift than the bottom 10% lowest-density patches and 3.5×\times larger teacher-masking answer contribution. Uniform weighting devotes many loss terms to low-information background patches, whereas sparse answer-bearing patches receive no special protection. The required intervention is minimal: we replace uniform residual alignment with density-weighted residual alignment, using patch self-dissimilarity as a training-free proxy for position importance. We call this HEED. Compared with normal end-to-end distillation, HEED increases performance by 8.7 points on OCRBench v2 and 5.13 points on a 10-benchmark average. The gain is realized on different teacher models and hybrid architectures. After standard post-training, the student reaches teacher-level performance on the 10-benchmark average with a 4.12×\times throughput and a 68% memory saving at 128k context, with no additional parameters and no inference-time cost.

Yihao Liang, Niraj K. Jha

Princeton University

Cross-Modal Knowledge DistillationEfficient VLM InferenceVLM DistillationCross-Architecture Knowledge DistillationTeacher-Student LearningKnowledge Distillation
AskPDF
May 15, 2026

GeoWorld-VLM: Geometry from World Models for Vision-Language Models

Modern Vision-Language Models (VLMs) achieve strong semantic recognition, yet remain brittle on elementary spatial relations such as left of, on, behind, and between. One cause of this failure arises before language reasoning begins: the visual pathway may compress or discard critical 3D structural cues during feature extraction, so the language model receives image representations that are already insufficient for reliable spatial judgment. We introduce GeoWorld-VLM, a VLM-side distillation framework that transfers geometric structure from frozen camera-conditioned video world models into VLMs. GeoWorld-VLM fine-tunes only the image encoder and multimodal projector, aligning post-projector image features with intermediate world-model representations while leaving the main backbone frozen. Given images, a prompt, and a sampled camera trajectory, the world-model teacher converts static visual input into a synthetic multi-view spatial signal. Training combines spatial answer supervision, teacher-student feature alignment, and a preservation anchor to the original VLM. Since the language model remains frozen, GeoWorld-VLM preserves the original model's linguistic capabilities while attributing spatial improvements to the enhanced visual pathway. To evaluate the effectiveness and generality of the proposed method, we apply GeoWorld-VLM to two distinct VLM architectures and observe consistent improvements across both backbones. GeoWorld-VLM improves performance by approximately 4 percent on both the What'sUp and VSR benchmarks, suggesting that world-model-guided visual alignment generalizes across model structures and spatial reasoning datasets.

Renjie Gu, Kaichen Zhou, Yan Luo +1

Harvard AI and Robotics Lab · Kempner Institute for the Study of Natural and Artificial Intelligence · Harvard University

Visual Spatial ReasoningVLM AdaptationVLM DistillationVisually Grounded ReasoningVision-Language AlignmentKnowledge Distillation
AskPDF
May 15, 2026

Offline Semantic Guidance for Efficient Vision-Language-Action Policy Distillation

Billion-parameter Vision-Language-Action (VLA) policies have recently shown impressive performance in robotic manipulation, yet their size and inference cost remain major obstacles for real-time closed-loop control. We introduce VLA-AD, a distillation framework that uses a Vision-Language Model as an offline semantic supervisor to transfer large VLA teachers into lightweight student policies. Instead of relying only on low-level action imitation, VLA-AD augments teacher-provided 7-DoF action targets with high-level semantic guidance, including task phase anchors and multi-frame operating-direction descriptions. These auxiliary signals are used only during training: at test time, the student policy runs independently, with neither the VLA teacher nor the VLM required. We evaluate VLA-AD on three LIBERO benchmark suites. Using OpenVLA-7B as the teacher, our method produces a 158M-parameter student, yielding a 44×44\times reduction in model size while matching the teacher with only a 0.27%0.27\% average relative gap. The resulting policy runs at 12.5 Hz on an RTX 4090, achieving a 3.28×3.28\times inference speedup over OpenVLA-7B. We further show that the same semantic distillation pipeline generalizes to a different π0.5π_{0.5}-4B teacher, where the student outperforms the teacher on two suites and remains within 0.53%0.53\% on libero\_goal. Additional analysis indicates that phase-level supervision and multi-frame directional cues make the student less sensitive to noisy teacher actions, such as erroneous high-frequency gripper changes. Overall, VLA-AD demonstrates that offline semantic guidance from VLMs can substantially improve the efficiency, robustness, and deployability of VLA policy distillation.

Jin Shi, Brady Zhang, Yishun Lu

Department of Mechanical Engineering University College London London, United Kingdom · Department of Engineering Science University of Oxford Oxford, United Kingdom

VLM DistillationEfficient VLA Model InferenceVisuomotor Policy LearningEfficient VLA ModelsRobotic ManipulationVision-Language-Action Models+1
AskPDF
May 15, 2026

DeltaPrompts: Escaping the Zero-Delta Trap in Multimodal Distillation

Distillation enables compact Vision-Language Models (VLMs) to obtain strong reasoning capabilities, yet the prompts driving this process are typically chosen via simple heuristics or aggregated from off-the-shelf datasets. We reveal a critical inefficiency in this approach: up to 69% of the prompts in standard chart / document reasoning datasets are effectively zero-delta, meaning the teacher and student already induce the exact same answer distribution. Training on these prompts provides minimal learning signal, causing student improvement to rapidly saturate regardless of data scale. To escape the zero-delta trap, we return to first principles: distillation fundamentally minimizes distributional divergence, and thus a prompt is valuable only if it exposes a functional capability gap between the teacher and student. We quantify this gap through answer divergence (ΔΔ), demonstrating that non-zero divergence is critical for effective scaling. Building on this insight, we propose a staged synthesis pipeline that repurposes existing datasets as seeds, actively targeting student failure modes to produce better prompts. The result is DeltaPrompts, a diverse dataset of 200k synthetic, high-divergence reasoning problems. We evaluate DeltaPrompts across three distinct settings: on-policy distillation with the target teacher-student pair, transfer to a novel model family without regenerating the data, and off-policy fine-tuning of a non-reasoning model. Across all scenarios, DeltaPrompts drives substantial gains, yielding up to 15% relative improvement even on top of a highly-optimized reasoning model (e.g., Qwen3-VL-8B-Thinking) -- averaged over 10 benchmarks spanning chart, document and perception-centric reasoning.

Jaehun Jung, Hyunwoo Kim, Brandon Cui +4

NVIDIA Research

VLM DistillationVLM ReasoningOn-Policy DistillationTeacher-Student LearningKnowledge Distillation
AskPDF
May 12, 2026

Hide to See: Reasoning-prefix Masking for Visual-anchored Thinking in VLM Distillation

Recent think-answer approaches in VLMs, such as Qwen3-VL-Thinking, boost reasoning performance by leveraging intermediate thinking steps before the final answer, but their computational cost becomes substantial, especially for larger VLMs. To distill such capabilities into compact think-answer VLMs, a primary objective is to improve the student's ability to utilize visual evidence throughout its reasoning trace, as long think-answer traces suffer from visual forgetting issues. To this end, we introduce a novel think-answer distillation framework that encourages the student to anchor its thinking on visual information by masking the student's salient reasoning prefixes. To compensate for such masked textual cues, the student is encouraged to rely more on visual evidence as an alternative source of information during distillation. Our masking strategies include: 1) token-wise salient reasoning-prefix masking, which masks high-influence reasoning prefixes selectively for each next-token prediction, and 2) self-paced masking budget scheduling, which gradually increases the masking scale according to distillation difficulty, measured by the discrepancy between teacher--student distributions. In the distillation phase, the student is guided by our salient reasoning-prefix mask, which blocks both future tokens and salient reasoning cues, in place of the standard causal mask used for auto-regressive language modeling. Experimental results show that our approach outperforms recent open-source VLMs, VLM distillation, and self-distillation methods on multimodal reasoning benchmarks, while further analyzes confirm enhanced visual utilization along the student thinking process.

Seonghoon Yu, Dongjun Nam, Byung-Kwan Lee +1

1KAIST · 3POSTECH · 2NVIDIA

Cross-Modal Knowledge DistillationCoT DistillationVLM DistillationVLM Reasoning
AskPDF
May 12, 2026

OmniThoughtVis: A Scalable Distillation Pipeline for Deployable Multimodal Reasoning Models

Recent multimodal large language models (MLLMs) have shown strong chain-of-thought (CoT) reasoning ability on vision-language tasks, but their direct deployment in real-world systems is often limited by latency and resource constraints. In practice, smaller MLLMs are preferred for online serving, yet their reasoning performance is bottlenecked by the lack of large-scale, high-quality multimodal CoT supervision. In this paper, we present OmniThoughtVis, a scalable data curation and distillation pipeline for transferring multimodal reasoning capabilities from high-capacity teacher models to smaller, deployment-oriented MLLMs. Starting from a diverse open-source seed pool, our pipeline generates structured CoT traces and performs joint annotation of reasoning difficulty, answer quality, and semantic task tags. To maintain data quality at scale, we combine rule-based filtering, difficulty-aware selection, and tag-based diversity sampling, resulting in a curated corpus of 1.8M samples that supports controllable subset construction for downstream training. We use OmniThoughtVis to distill Qwen3-VL models from 2B to 8B parameters and evaluate them on nine multimodal reasoning benchmarks. The resulting distilled models show consistent gains across model scales, including improvements of up to +16.8 points on MathVerse and +5.6 points on MMMU-Pro for the 4B model. Notably, the distilled 4B model matches or surpasses the undistilled 8B baseline on several tasks, highlighting the practical value of scalable reasoning distillation for deployment-oriented MLLMs.

Yuanhao Yue, Chengyu Wang, Yuanjie Lyu +2

Alibaba Group, Hangzhou, China

Multimodal CoT ReasoningMultimodal ReasoningLanguage Model DistillationVLM DistillationMultimodal UnderstandingKnowledge Distillation+1
AskPDF
May 11, 2026

LLaVA-CKD: Bottom-Up Cascaded Knowledge Distillation for Vision-Language Models

Large Vision-Language Models (VLMs) are successful in addressing a multitude of vision-language understanding tasks, such as Visual Question Answering (VQA), but their memory and compute requirements remain a concern for practical deployment. A promising class of techniques for mitigating this concern is Knowledge Distillation, where knowledge from a high-capacity Teacher network is transferred to a considerably smaller Student network. However, the capacity gap between the two networks is both a blessing and a curse: the smaller the Student network, the better its efficiency, and the stronger (typically, also larger) the Teacher, the more knowledge it carries; yet, beyond a point, the larger capacity gap between the two leads to worse knowledge transfer. To counter this effect, we propose a bottom-up cascaded knowledge distillation (CKD) framework. Instead of treating knowledge transfer as an activity involving one high-capacity Teacher (or an ensemble of such), inspired by human formal education systems, we introduce one (potentially, more) additional Teacher(s) of intermediate capacity that gradually bring the Student network to the next level, where the next (higher-capacity) Teacher can take over. We provide a theoretical analysis to motivate why progressive reduction of the Teacher-Student gap may improve distillation. We apply the proposed framework on models built upon the LLaVA methodology and evaluate the derived models on seven standard, publicly available VQA benchmarks, demonstrating competitive performance with recent compact VLMs. Source code for cascaded distillation and trained models are available at https://github.com/IDT-ITI/LLaVA-CKD.

Nikolaos Gkalelis, Vasileios Mezaris

CERTH-ITI, Thessaloniki, Greece, 57001

Vision-Language ModelsEfficient VLM InferenceVLM DistillationMulti-Teacher Knowledge DistillationKnowledge Distillation
AskPDF
May 10, 2026

Distilling 3D Spatial Reasoning into a Lightweight Vision-Language Model with CoT

Large-scale 3D vision-language models (VLMs) like LLaVA-3D offer strong spatial reasoning but are difficult to deploy due to high computational costs. We propose a knowledge distillation framework that transfers spatial reasoning from a 7B teacher to a 2.29B student model. Our approach achieves 8.7x lower inference latency and a 3x reduction in model size while retaining 54-72% of the teacher's performance. The framework utilizes VGGT as the vision encoder and a multi-task distillation pipeline with uncertainty-aware loss weighting. To improve reasoning without chain-of-thought (CoT) data, we introduce "Hidden CoT": learnable latent tokens that serve as an internal scratchpad before answer generation. This is the first use of latent scratchpad reasoning in distilled 3D VLMs. The student model jointly performs spatial description, depth estimation, and object detection. Experiments on ScanNet and 3D-FRONT show strong spatial understanding, reaching 68-72% accuracy on proximity and contact tasks. Our framework enables efficient 3D scene QA on resource-constrained platforms.

Alaa Asfour, Christopher Indris, Leihan Chen +2

Department of Computer Science, Toronto Metropolitan University, Toronto, ON, Canada

Cross-Modal Knowledge Distillation3D Spatial ReasoningEfficient VLM InferenceVLM Distillation3D Scene Understanding3D VLMs
AskPDF
Apr 16, 2026

Switch-KD: Visual-Switch Knowledge Distillation for Vision-Language Models

Vision-Language Models (VLMs) have shown remarkable capabilities in joint vision-language understanding, but their large scale poses significant challenges for deployment in resource-constrained scenarios. Knowledge Distillation (KD) offers a viable way to improve model capabilities without increasing model size or data requirements, making deployment more efficient. However, applying KD to VLMs is challenged by modality-specific supervision: although multimodal knowledge in VLMs is fused within the language space, current methods supervise each modality separately without explicitly addressing multimodal alignment, leading to inconsistent multimodal knowledge transfer. To address this, we propose Switch-KD, a visual-switch distillation framework that unifies vision-language knowledge transfer within a shared text-probability space. Switch-KD comprises two key components: (1) Visual-Switch Distillation, which switches the student's visual outputs into the teacher's language pathway to construct cross-modal probabilistic references for implicit visual knowledge transfer; and (2) Dynamic Bi-directional Logits Difference (DBiLD) loss, which adaptively aligns informative probability regions while preserving the distributional structures of teacher and student through bidirectional supervision. Guided by Switch-KD, a 0.5B TinyLLaVA effectively distills rich multimodal knowledge from its 3B teacher, yielding an average improvement of 3.6 points across 10 multimodal benchmarks without any architectural modification.

Haoyi Sun, Xiaoxiao Wang, Ning Mao +5

Li Auto Inc.

Cross-Modal Knowledge DistillationVision-Language ModelsVLM DistillationVision-Language AlignmentKnowledge Distillation
AskPDF
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 ModelsDiffusion VLMsVLM AdaptationDiffusion Model DistillationEfficient VLM InferenceVLM Distillation+1
AskPDF
Apr 13, 2026

TIPSv2: Advancing Vision-Language Pretraining with Enhanced Patch-Text Alignment

Recent progress in vision-language pretraining has enabled significant improvements to many downstream computer vision applications, such as classification, retrieval, segmentation and depth prediction. However, a fundamental capability that these models still struggle with is aligning dense patch representations with text embeddings of corresponding concepts. In this work, we investigate this critical issue and propose novel techniques to enhance this capability in foundational vision-language models. First, we reveal that a patch-level distillation procedure significantly boosts dense patch-text alignment -- surprisingly, the patch-text alignment of the distilled student model strongly surpasses that of the teacher model. This observation inspires us to consider modifications to pretraining recipes, leading us to propose iBOT++, an upgrade to the commonly-used iBOT masked image objective, where unmasked tokens also contribute directly to the loss. This dramatically enhances patch-text alignment of pretrained models. Additionally, to improve vision-language pretraining efficiency and effectiveness, we modify the exponential moving average setup in the learning recipe, and introduce a caption sampling strategy to benefit from synthetic captions at different granularities. Combining these components, we develop TIPSv2, a new family of image-text encoder models suitable for a wide range of downstream applications. Through comprehensive experiments on 9 tasks and 20 datasets, we demonstrate strong performance, generally on par with or better than recent vision encoder models. Code and models are released via our project page at https://gdm-tipsv2.github.io/ .

Bingyi Cao, Koert Chen, Kevis-Kokitsi Maninis +16

Google DeepMind · xAI · Epsilon Health +1

Vision-Language PretrainingVision-Language AlignmentMasked Image ModelingVLM Distillation
AskPDF
Apr 9, 2026

Orion-Lite: Distilling LLM Reasoning into Efficient Vision-Only Driving Models

Leveraging the general world knowledge of Large Language Models (LLMs) holds significant promise for improving the ability of autonomous driving systems to handle rare and complex scenarios. While integrating LLMs into Vision-Language-Action (VLA) models has yielded state-of-the-art performance, their massive parameter counts pose severe challenges for latency-sensitive and energy-efficient deployment. Distilling LLM knowledge into a compact driving model offers a compelling solution to retain these reasoning capabilities while maintaining a manageable computational footprint. Although previous works have demonstrated the efficacy of distillation, these efforts have primarily focused on relatively simple scenarios and open-loop evaluations. Therefore, in this work, we investigate LLM distillation in more complex, interactive scenarios under closed-loop evaluation. We demonstrate that through a combination of latent feature distillation and ground-truth trajectory supervision, an efficient vision-only student model Orion-Lite can even surpass the performance of its massive VLA teacher, ORION. Setting a new state-of-the-art on the rigorous Bench2Drive benchmark, with a Driving Score of 80.6. Ultimately, this reveals that vision-only architectures still possess significant, untapped potential for high-performance reactive planning.

Jing Gu, Niccolò Cavagnero, Gijs Dubbelman

Eindhoven University of Technology

End-to-End Autonomous DrivingVLM DistillationKnowledge DistillationReasoning for VLA Models
AskPDF
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 AdaptationVLM DistillationCross-Modal Knowledge DistillationContinual Learning for VLMs
AskPDF
Mar 30, 2026

MEDiC: Multi-objective Exploration of Distillation from CLIP

Masked image modeling (MIM) methods typically operate in either raw pixel space (reconstructing masked patches) or latent feature space (aligning with a pre-trained teacher). We present MEDiC (Multi-objective Exploration of Distillation from CLIP), a framework that combines both spaces in a single pipeline through three complementary objectives: patch-level token distillation from a frozen CLIP encoder, global CLS alignment, and pixel reconstruction via a lightweight decoder. We conduct a systematic investigation of the design space surrounding this multi-objective framework. First, we show that all three objectives provide complementary information, with the full combination reaching 73.9% kNN accuracy on ImageNet-1K. Second, we introduce hierarchical clustering with relative position bias for evolved masking and find that, despite producing more semantically coherent masks than prior methods, evolved masking does not outperform simple block masking in the teacher-guided distillation setting, a finding we attribute to the teacher's inherent semantic awareness. Third, we reveal that optimal scalar loss weights are extremely fragile, with small perturbations causing drops of up to 17 percentage points in kNN accuracy. Our framework achieves 73.9% kNN and 85.1% fine-tuning accuracy with ViT-Base at 300 epochs.

Konstantinos Georgiou, Maofeng Tang, Hairong Qi

Min H. Kao Department of Electrical Engineering and Computer Science The University of Tennessee Knoxville, TN 37996

Masked Image ModelingVLM DistillationFeature-Based Knowledge DistillationImage Classification
AskPDF
Mar 26, 2026

CLIP-RD: Relational Distillation for Efficient CLIP Knowledge Distillation

Contrastive Language-Image Pre-training (CLIP) demonstrates strong zero-shot generalization, but due to substantial computational and memory costs, distillation into lightweight models is required. Existing relational objectives do not explicitly model multidirectional relationships between teacher and student embeddings, potentially leaving the geometric relationships insufficiently constrained. This may disrupt the modality-gap structure important for zero-shot transfer. To address these limitations, we propose a relational distillation framework, CLIP-RD, which introduces two relational methods, Vertical Relational Distillation (VRD) and Cross Relational Distillation (XRD). VRD aligns teacher-student intra-modal similarity distributions to enforce consistent distillation strength across image and text embeddings. Meanwhile, XRD aligns the teacher-image-student-text and teacher-text-student-image similarity distributions to impose bidirectional cross-modal symmetry. By jointly modeling these multidirectional relational structures, CLIP-RD aligns the student's embedding geometry more faithfully to the teacher's, outperforming CLIP-KD by 1.8%p. This performance improvement is maintained across diverse architectures, teacher scales, retrieval tasks, downstream tasks, and corruption settings, with negligible additional training-time overhead.

Jeannie Chung, Hanna Jang, Ingyeong Yang +2

Ewha Womans University, Seoul 03760, South Korea

Cross-Modal LearningCross-Modal Knowledge DistillationVision-Language ModelsVLM DistillationRelational Knowledge DistillationKnowledge Distillation
AskPDF
Mar 25, 2026

GoldiCLIP: The Goldilocks Approach for Balancing Explicit Supervision for Language-Image Pretraining

Until recently, the success of large-scale vision-language models (VLMs) has primarily relied on billion-sample datasets, posing a significant barrier to progress. Latest works have begun to close this gap by improving supervision quality, but each addresses only a subset of the weaknesses in contrastive pretraining. We present GoldiCLIP, a framework built on a Goldilocks principle of finding the right balance of supervision signals. Our multifaceted training framework synergistically combines three key innovations: (1) a text-conditioned self-distillation method to align both text-agnostic and text-conditioned features; (2) an encoder integrated decoder with Visual Question Answering (VQA) objective that enables the encoder to generalize beyond the caption-like queries; and (3) an uncertainty-based weighting mechanism that automatically balances all heterogeneous losses. Trained on just 30 million images, 300x less data than leading methods, GoldiCLIP achieves state-of-the-art among data-efficient approaches, improving over the best comparable baseline by 2.2 points on MSCOCO retrieval, 2.0 on fine-grained retrieval, and 5.9 on question-based retrieval, while remaining competitive with billion-scale models. Project page: https://petsi.uk/goldiclip.

Deen Dayal Mohan, Hossein Souri, Vitali Petsiuk +4

AI Center – Mountain View, Samsung Electronics

Vision-Language PretrainingVLM DistillationAdaptive Loss Weighting
AskPDF
Mar 24, 2026

Adaptive Dual-Teacher Distillation with Subnetwork Rectification for Bridging Semantic Gaps in Black-Box Domain Adaptation

Assuming that neither source data nor source model parameters are accessible, black-box domain adaptation (BBDA) represents a highly practical yet challenging setting, where transferable knowledge is limited to the predictions of a black-box source model. Existing approaches exploit such knowledge via pseudo-label refinement or by leveraging vision-language models (ViLs), but they often fail to reconcile the inherent discrepancy between task-specific knowledge from black-box models and language-aligned semantic priors of ViLs, resulting in suboptimal integration and degraded adaptation performance. To address this challenge, we propose adaptive Dual-Teacher Distillation with Subnetwork Rectification (DDSR), a framework that explicitly reconciles these complementary yet inconsistent knowledge sources. DDSR employs an adaptive prediction fusion strategy to integrate predictions from the black-box source model and a ViL, generating reliable pseudo-labels for the target domain. A subnetwork-based regularization mechanism mitigates overfitting to noisy supervision by enforcing output consistency and gradient divergency. Furthermore, progressively improved target predictions iteratively refine both pseudo-labels and ViL prompts, enhancing semantic alignment. Finally, class-wise prototypes are used to further optimize target predictions via self-training. Extensive experiments on multiple benchmark datasets demonstrate that DDSR consistently outperforms state-of-the-art methods, including those with access to source data or source model parameters.

Zhe Zhang, Jing Li, Wanli Xue +4

School of Computer Science and Engineering, Tianjin University of Technology, Tianjin 300384, China. · School of Artificial Intelligence, Tianjin University, Tianjin 300350, China

Source-Free Domain AdaptationKnowledge DistillationVLM DistillationVLM Adaptation
AskPDF
Mar 22, 2026

Uncertainty-Aware Knowledge Distillation for Multimodal Large Language Models

Knowledge distillation establishes a learning paradigm that leverages both data supervision and teacher guidance. However, determining the optimal balance between learning from data and learning from the teacher is challenging, as some samples may be noisy while others are subject to teacher uncertainty. This motivates the need for adaptively balancing data and teacher supervision. We propose Beta-weighted Knowledge Distillation (Beta-KD), an uncertainty-aware distillation framework that adaptively modulates how much the student relies on teacher guidance. Specifically, we formulate teacher--student learning from a unified Bayesian perspective and interpret teacher supervision as a Gibbs prior over student activations. This yields a closed-form, uncertainty-aware weighting mechanism and supports arbitrary distillation objectives and their combinations. Extensive experiments on multimodal VQA benchmarks demonstrate that distilling student Vision-Language Models from a large teacher VLM consistently improves performance. The results show that Beta-KD outperforms existing knowledge distillation methods. The code is available at https://github.com/Jingchensun/beta-kd.

Jingchen Sun, Shaobo Han, Deep Patel +3

NEC Laboratories America, Inc., USA · University at Buffalo, SUNY · Rutgers University

Uncertainty-Aware Knowledge DistillationKnowledge DistillationVLM DistillationMultimodal Large Language Models
AskPDF
Mar 21, 2026

VSD-MOT: End-to-End Multi-Object Tracking in Low-Quality Video Scenes Guided by Visual Semantic Distillation

Existing multi-object tracking algorithms typically fail to adequately address the issues in low-quality videos, resulting in a significant decline in tracking performance when image quality deteriorates in real-world scenarios. This performance degradation is primarily due to the algorithms' inability to effectively tackle the problems caused by information loss in low-quality images. To address the challenges of low-quality video scenarios, inspired by vision-language models, we propose a multi-object tracking framework guided by visual semantic distillation (VSD-MOT). Specifically, we introduce the CLIP Image Encoder to extract global visual semantic information from images to compensate for the loss of information in low-quality images. However, direct integration can substantially impact the efficiency of the multi-object tracking algorithm. Therefore, this paper proposes to extract visual semantic information from images through knowledge distillation. This method adopts a teacher-student learning framework, with the CLIP Image Encoder serving as the teacher model. To enable the student model to acquire the capability of extracting visual semantic information suitable for multi-object tracking tasks from the teacher model, we have designed the Dual-Constraint Semantic Distillation method (DCSD). Furthermore, to address the dynamic variation of frame quality in low-quality videos, we propose the Dynamic Semantic Weight Regulation (DSWR) module, which adaptively allocates fusion weights based on real-time frame quality assessment. Extensive experiments demonstrate the effectiveness and superiority of the proposed method in low-quality video scenarios in the real world. Meanwhile, our method can maintain good performance in conventional scenarios.

Jun Du

Multi-Object TrackingVisual Object TrackingKnowledge DistillationVLM Distillation
AskPDF
Mar 16, 2026

DAIT: Distillation from Vision-Language Models to Lightweight Classifiers with Adaptive Intermediate Teacher Transfer

Large-scale Vision-Language Models (VLMs) encode rich multimodal semantics that are highly beneficial for fine-grained visual categorization (FGVC). However, their prohibitive computational cost hinders practical deployment in resource-constrained environments. Although knowledge distillation contributes to transferring VLMs capacity to lightweight classifiers, conventional distillation mechanisms, which directly transfer from a generic VLM to a compact student, often yield suboptimal results due to severe architectural misalignment and introducing task-irrelevant information. To alleviate this limitation, we propose Distillation with Adaptive Intermediate Teacher transfer (DAIT) in this study, facilitating adaptive knowledge transfer from VLMs to lightweight students. DAIT introduces a trainable intermediate teacher that learns to transfer frozen VLMs representations under explicit supervision from the target fine-grained task. This intermediate teacher adaptively enhances discriminative visual cues, thereby producing compact and task-aligned knowledge that can be reliably distilled into lightweight models. Extensive evaluations on multiple FGVC benchmarks with diverse student architectures demonstrate that our method achieves respective performance gains of 12.63% and 8.34% on FGVC-Aircraft and CUB-200-2011 datasets, establishing DAIT as a principled paradigm for transferring from general-purpose VLMS to deployable fine-grained recognition models.

Zhengxu He, Jun Li, Zhijian Wu

School of Computer and Electronic Information, Nanjing Normal University, Nanjing, China · Medical Artificial Intelligence Lab, Westlake University

VLM DistillationFine-Grained Image ClassificationFeature-Based Knowledge Distillation
AskPDF
Mar 13, 2026

NanoVDR: Distilling a 2B Vision-Language Retriever into a 70M Text-Only Encoder for Visual Document Retrieval

Vision-Language Model (VLM) based retrievers have advanced visual document retrieval (VDR) to impressive quality. They require the same multi-billion parameter encoder for both document indexing and query encoding, incurring high latency and GPU dependence even for plain-text queries. We observe that this design is unnecessarily symmetric: documents are visually complex and demand strong visual understanding, whereas queries are just short text strings. NanoVDR exploits this query--document asymmetry by decoupling the two encoding paths: a frozen 2B VLM teacher indexes documents offline, while a distilled text-only student as small as 69M parameters encodes queries at inference. The key design choice is the distillation objective. Through systematic comparison of six objectives across three backbones and 22 ViDoRe benchmark datasets, we find that pointwise cosine alignment on query text consistently outperforms ranking-based and contrastive alternatives, while requiring only pre-cached teacher query embeddings and no document processing during training. Furthermore, we identify cross-lingual transfer as the primary performance bottleneck, and resolve it cheaply by augmenting training data with machine-translated queries. The resulting NanoVDR-S-Multi (DistilBERT, 69M) retains 95.1% of teacher quality and outperforms DSE-Qwen2 (2B) on v2 and v3 with 32×\times fewer parameters and 50×\times lower CPU query latency, at a total training cost under 13 GPU-hours.

Zhuchenyang Liu, Yao Zhang, Yu Xiao

Aalto University, Espoo, Finland

Visual Document RetrievalVision-Language RetrievalVLM DistillationEfficient VLM Inference
AskPDF
Mar 12, 2026

Beyond Single-Sample: Reliable Multi-Sample Distillation for Video Understanding

Traditional black-box distillation for Large Vision-Language Models (LVLMs) typically relies on a single teacher response per input, which often yields high-variance responses and format inconsistencies in multimodal or temporal scenarios. To mitigate this unreliable supervision, we propose R-MSD (Reliable Multi-Sample Distillation), a framework that explicitly models teacher sampling variance to enhance distillation stability. Rather than relying on a single teacher response, our approach leverages a task-adaptive teacher pool to provide robust supervision tailored to both closed-ended and open-ended reasoning. By integrating quality-aware signal matching with an adversarial distillation objective, our approach effectively filters teacher noise while maximizing knowledge transfer. Extensive evaluations across comprehensive video understanding benchmarks demonstrate that R-MSD consistently outperforms single sample distillation methods. We additionally include an original SFT+RL 4B baseline under the same training budget, which shows only marginal gains, while our method achieves significant improvements. With a 4B student model, our approach delivers gains on VideoMME (+1.5%), Video-MMMU (+3.2%), and MathVerse (+3.6%).

Songlin Li, Xin Zhu, Zechao Guan +2

University of Electronic Science and Technology of China · Robotics Center, XPeng Motors

VLM DistillationKnowledge DistillationUncertainty-Aware Knowledge DistillationMulti-Teacher Knowledge DistillationVideo Understanding
AskPDF
Mar 11, 2026

From Images to Words: Efficient Cross-Modal Knowledge Distillation to Language Models from Black-box Teachers

Knowledge distillation (KD) methods are pivotal in compressing large pre-trained language models into smaller models, ensuring computational efficiency without significantly dropping performance. Traditional KD techniques assume homogeneity in modalities between the teacher (source) and the student (target) models. On the other hand, existing multimodal knowledge distillation methods require modality-specific pre-training of the teacher model, which is computationally infeasible in most cases. In this paper, we introduce ARMADA, an efficient cross-modal knowledge distillation framework designed to transfer knowledge from large vision-language models, including black-box models, to language-only models. Unlike existing KD techniques that rely on the internal structures of multimodal teachers or require computationally expensive pre-training, ARMADA leverages novel alignment techniques to distil knowledge without altering the teacher model, ensuring efficiency and scalability. We empirically validate ARMADA on twelve natural language understanding, eight complex generative reasoning and five instruction-tuning tasks, demonstrating consistent performance improvements in large models such as DeBERTa-v2-1.4B, OPT-1.3B, LLaMA-{3B, 7B, 8B}. ARMADA achieves up to 3.4% improvement on language understanding tasks and 2.6% boost in generative reasoning, all without requiring expensive multimodal pre-training or fine-tuning of the teacher model. Our findings challenge conventional knowledge distillation paradigms by demonstrating that even vision-language models, despite lacking direct textual understanding, can significantly enhance language models when distilled appropriately.

Ayan Sengupta, Shantanu Dixit, Md Shad Akhtar +1

Indian Institute of Technology Delhi, India · Indraprastha Institute of Information Technology Delhi, India

Cross-Modal Knowledge DistillationKnowledge DistillationLanguage Model DistillationVLM Distillation
AskPDF