arXiv subjects

Jungbeom Lee

Publications and source records attributed to Jungbeom Lee.

At least 19 recordsLinked to original sources

arXiv2026

Heads That Write, Not Just Point: Image Retrieval Heads in Vision-Language Models

Large vision-language models (LVLMs) can process interleaved text and multiple images in a single context, but how they internally identify the image relevant to a question remains poorly understood. We study this grounding step as in-context image retrieval and introduce ICIR-MCQ, a controlled probe for head-level analysis of image retrieval. Using each head's attention mass over image-token spans, we first identify sparse pointer heads that reliably point to the target image. These heads transfer beyond the probe: a single pointer head is competitive with external vision-language retrievers on a naturalistic multi-image retrieval benchmark, showing that LVLMs contain a strong internal image retrieval signal. Attention alone, however, shows where a head reads from, not what its output passes on to later layers, so we introduce a complementary output-side analysis. For each head, we train a lightweight image retriever on its value-weighted output and use it to identify writer heads, whose outputs contain enough information to identify the target image. Surprisingly, the pointer and writer criteria select substantially different head sets, with only 35 to 47 heads overlapping among the top-100 heads under each criterion. Targeted set-partition ablations show that the heads most important for standard multi-image inference are concentrated in this pointer-writer intersection, while heads selected by only one criterion show less consistent separation from random-head controls. The most consistent causal effects arise from heads selected by both the attention-based pointer and output-side writer criteria. Code and data: https://github.com/parkquasar/Image-Retrieval-Heads.

cs.CV↗

arXiv2026

V-LUMEN: Visual Lookup Memory for Embedding Scaling in Vision-Language Models

Embedding-level scaling expands the capacity of language models through lookup-based memory rather than by increasing model depth, width, or dense computation. Extending this idea to vision-language models (VLMs) is challenging because visual features are continuous, spatially structured, and query-dependent, unlike discrete text tokens that naturally support n-gram lookup. We introduce V-LUMEN, a visual lookup memory for VLMs. V-LUMEN discretizes visual features with a frozen codebook, builds spatially aware memory keys through spatial aggregation-based hashing, and modulates them with text-conditioned relevance signals; all addressing is non-parametric, and the retrieved embeddings are fused into the backbone through a gated residual pathway. On five reasoning-intensive benchmarks, V-LUMEN substantially improves a frozen Qwen3-VL-2B backbone and transfers zero-shot to three benchmarks whose images are disjoint from the training data. At a matched parameter budget, moving part of a LoRA or MoE adapter's capacity into V-LUMEN preserves accuracy on two backbones, indicating that external visual memory is an independent capacity axis rather than a stronger adaptation method. Because lookup cost is constant in table size and incurred only once at prefill, this capacity grows in storage rather than in per-token compute. Code is available at https://github.com/miso-choi/V-Lumen.

cs.LG↗

arXiv2026

RECAST: Recasting Vision-Language Semantics into an Actionable Cost Map for Robot Navigation

Safe and robust robot navigation across diverse environments requires a high-level understanding of complex scenes and the ability to carry it into stable motion. Recent works tackle this with learning-based models trained at scale and with approaches built on vision-language models (VLMs). However, learning-based models break down outside their training distribution, while VLM-based approaches bring that understanding but rarely ground it in the scene or align the action with it. To address these limitations, we present RECAST, a robot navigation framework that combines the reasoning of a VLM with the spatial grounding of vision foundation models to build an Actionable Cost map. Given the robot's front view and the user's instruction, we first decompose the scene with the VLM, judging which surfaces are traversable, which objects pose a risk, which heading to prefer, and which gaps are passable. Vision foundation models then ground these surfaces and objects in the image, and all four judgments are spatially recast into one compact cost map. This map both conditions the trajectory decoders and scores their proposals to select the one to execute. As the VLM's answers trail the live scene, both steps draw on cost maps from two points in time: the pivot frame the VLM judged, which carries all four judgments, and the current frame, whose terrain and collision costs are rebuilt from the current image. RECAST improves success over the strongest prior method by 13.3 points in simulation and 31.4 points on a real quadruped, and reduces the collision rate relative to it by 9.6 and 14.3 points, reaching the lowest collision rate among all methods. The project page is available at https://recast-nav.github.io/

cs.RO↗

arXiv2026

SAGE: Semantic Attribute Graphs for Multi-Entity Visual Retrieval

Dense document images often contain many fine-grained visual and textual entities whose relevance depends on a user query. Standard vision-language retrievers encode cropped regions with a single vector, which can mix distinct entity signals and obscure the evidence needed for fine-grained retrieval. We call this failure mode Semantic Dilution and quantitatively show that it degrades entity-level retrieval as a function of entity density. To mitigate it, we propose SAGE, a training-free framework that parses semantic entities from dense document images, represents them as hierarchical graph nodes with multi-vector embeddings, and retrieves query-relevant evidence through iterative entity-level subgraph matching. We also introduce DEAR, a dataset of 1,055 query--image pairs sourced from product detail pages, where each query requires retrieving and comparing multiple fine-grained entities from visually dense inputs across four question types of increasing complexity. Experiments show that SAGE substantially reduces semantic dilution and outperforms patch-level and OCR-based retrieval baselines on DEAR, achieving a Recall@3 of 0.849 and a generation score of 2.746 on multi-entity visual comparison queries. Our code is available at https://github.com/All4Nothing/SAGE.

cs.IR↗

arXiv2026

The Truth Stays in the Family: Enhancing Contextual Grounding via Inherited Truthful Heads in Model Lineages

Recent advances in large language models (LLMs) have produced many specialized multimodal LLMs (MLLMs) that share common foundational LLMs, forming distinct model lineages. It remains unclear whether a fundamental behavioral link exists between the foundational LLMs and downstream variants. We investigate this question by quantifying head-level context-truthfulness scores. Across diverse LLM and MLLM lineages, including Vicuna-, Qwen2.5-, LLaMA2-, and Mistral-based models, we find that Truth Scores are strongly preserved within model families, even after instruction tuning or multimodal adaptation. We further show that this inheritance is consistent with attention-head weight preservation, and that context-truthful heads attend to query-relevant evidence. Building on this finding, we propose TruthProbe, a soft-gating strategy that amplifies context-truthful heads while preserving other head contributions. TruthProbe improves contextual truthfulness on HaluEval and reduces multimodal hallucination on POPE and CHAIR, with base-LLM Truth Scores transferring effectively to their fine-tuned LLM and MLLM descendants. Code is available at https://github.com/miso-choi/TruthProbe.

cs.CL↗

arXiv2026

Diagnosing and Correcting Concept Omission in Multimodal Diffusion Transformers

Multimodal Diffusion Transformers (MM-DiTs) have achieved remarkable progress in text-to-image generation, yet they frequently suffer from concept omission, where specified objects or attributes fail to emerge in the generated image. By performing linear probing on text tokens, we demonstrate that text embeddings can distinguish a characteristic `omission signal' representing the absence of target concepts. Leveraging this insight, we propose Omission Signal Intervention (OSI), which amplifies the omission signal to actively catalyze the generation of missing concepts. Comprehensive experiments on FLUX.1-Dev and SD3.5-Medium demonstrate that OSI significantly alleviates concept omission even in extreme scenarios.

cs.CV↗

arXiv2026

SpatiO: Adaptive Test-Time Orchestration of Vision-Language Agents for Spatial Reasoning

Understanding visual scenes requires not only recognizing objects but also reasoning about their spatial relationships. Unlike general vision-language tasks, spatial reasoning requires integrating multiple inductive biases, such as 2D appearance cues, depth signals, and geometric constraints, whose reliability varies across contexts. This suggests that effective spatial reasoning requires spatial adaptability: the ability to flexibly coordinate different reasoning strategies depending on the input. However, most existing approaches rely on a single reasoning pipeline that implicitly learns a fixed spatial prior, limiting their ability to adapt under distribution changes. Multi-agent systems offer a promising alternative by aggregating diverse reasoning trajectories, but prior attempts in spatial reasoning primarily employ homogeneous agents, restricting the diversity of inductive biases they can leverage. In this work, we introduce SpatiO, a heterogeneous multi-agent framework for spatial reasoning that coordinates multiple vision-language specialists with complementary inductive biases. To enable effective collaboration, we propose Test-Time Orchestration (TTO), an calibration mechanism that dynamically evaluates and reweights agents based on their observed reliability during inference, without modifying model parameters. Extensive experiments on diverse spatial reasoning benchmarks, including 3DSRBench, STVQA-7k, CV-Bench, and Omni3D-Bench, demonstrate that SpatiO consistently improves spatial reasoning performance over both closed-source and open-source baselines. The project page is available at https://cy-h1329.github.io/spatio/.

cs.CV↗

arXiv2026

Streaming Dense Voxel Representations for 3D Occupancy Prediction

In this paper, we explore dense voxel streaming for accurate and efficient 3D occupancy prediction. While dense voxel representations offer fine-grained spatial details and streaming paradigm enables efficient temporal processing, naively combining the two introduces key challenges: (i) warping-induced distortions caused by interpolation used for temporal alignment, and (ii) degraded dynamic object representations due to motion misalignment and detail loss in image-to-voxel projection. To address these, we propose StreamOcc, a novel framework that utilizes two aggregation strategies. Specifically, it first refines propagated voxel features to reduce warping artifacts before temporal accumulation, and then selectively injects instance-level query features encoding dynamic-object semantics into the corresponding occupied voxel regions, preserving temporally consistent modeling while strengthening dynamic object representations. Unlocking effective dense voxel streaming, StreamOcc achieves state-of-the-art performance on SurroundOcc-benchmark and Occ3D-nuScenes under real-time constraints, outperforming the prior best methods by +1.3/2.5 and +1.5/2.0 in (overall/dynamic object) mIoU, respectively, while running at 83.3 ms per frame with only 2.8 GB of memory. The project page is available at https://moonseokha.github.io/StreamOcc/.

cs.CV↗

arXiv2026

AxisGuide: Grounding Robot Action Coordinate System in RGB Observations for Robust Visuomotor Manipulation

Visuomotor manipulation policies trained via large-scale behavior cloning have achieved strong semantic scene understanding, yet often fail to reliably execute correct low-level actions under distribution shifts. For example, even in a simple pickup task with identical scene layouts, camera viewpoints, and illumination, performance can degrade substantially when the object is placed at unseen locations. We argue that this gap arises from insufficient action understanding, namely the inability to interpret the robot's base-frame action coordinate system in image space. To address this issue, we introduce AxisGuide, a lightweight guidance method that bridges semantic scene understanding and action-coordinate interpretation. Using camera parameters and end-effector poses, AxisGuide renders the robot base-frame axes in each camera view and augments RGB observations with a small set of cue channels that explicitly visualize the meaning of the +x, +y, and +z motions in image space. Extensive evaluations in both the LIBERO simulation and real-world environments demonstrate that AxisGuide yields substantial performance gains and improved generalization, highlighting the effectiveness of explicit action-coordinate cues for learning reliable and transferable generalist visuomotor policies.

cs.RO↗

arXiv2026

Causality-Aware End-to-End Autonomous Driving via Ego-Centric Joint Scene Modeling

End-to-end autonomous driving, which bypasses traditional modular pipelines by directly predicting future trajectories from sensor inputs, has recently achieved substantial progress. However, existing methods often overlook the causal inter-dependencies in ego-vehicle planning, ignoring the reciprocal relations between the ego vehicle and surrounding agents. This causal oversight leads to inconsistent and unreliable trajectory predictions, especially in interaction-critical scenarios where ego decisions and neighboring agent behaviors must be reasoned about jointly. To address this limitation, we propose CaAD, a Causality-aware end-to-end Autonomous Driving framework that captures these dependencies within a shared latent scene representation. First, we propose an ego-centric joint-causal modeling module that builds on the marginal prediction branch, and learns causal dependencies between the ego vehicle and interaction-relevant agents. Second, we employ a causality-aware policy alignment stage implemented with joint-mode embeddings to align the stochastic ego policy with planning-oriented closed-loop feedback computed from surrounding traffic and map context. On the Bench2Drive and NAVSIM benchmarks, CaAD demonstrates strong closed-loop planning performance, achieving a Driving Score of 87.53 and Success Rate of 71.81 on Bench2Drive, and a PDMS of 91.1 on NAVSIM. The project page is available at https://moonseokha.github.io/CaAD/.

cs.RO↗

arXiv2026

SODE: Analyzing Social Dynamics in LLM Agents

As Large Language Models (LLMs) evolve into interactive agents, understanding their behavioral alignment within human social dynamics becomes essential. While behavioral game theory offers a framework to study these interactions, previous work has predominantly relied on outcome-based metrics such as average scores. This focus overlooks the mechanisms that facilitate sustainable cooperation, as identical scores can be derived from vastly different strategies. To bridge this gap, we introduce SODE (Social Dynamics Evaluation), a framework that evaluates LLM agents across three evolutionary dimensions: Direct Reciprocity for strategy adaptation, Indirect Reciprocity for reputation sensitivity, and Group Dynamics for cooperative resilience. Applying SODE reveals systematic divergences: instruction-tuned models often exhibit "passive compliance" that renders them vulnerable to exploitation, while reasoning models prioritize short-horizon optimization, destabilizing long-term cooperation. Notably, we demonstrate that a "long-horizon framing" can unlock reciprocal capabilities in reasoning models. Thus, SODE offers a systematic, mechanism-grounded benchmark for aligning AI agents with complex human social dynamics.

cs.MA↗

arXiv2026

Dynamic In-Group Persona Generation for Enhancing Human-AI Rapport

LLM-based chatbots are increasingly applied in interpersonal domains such as counseling and peer support, where establishing human-AI rapport is crucial yet remains challenging. In this work, we introduce a novel approach for conditioning LLMs with in-group personas, which (i) first identifies a user's primary concern and brief personal context (e.g., a computer science undergraduate worried about future career prospects), and (ii) generates a synthetic in-group persona that shares a similar primary concern while differing in background and narrative details, such as age or profession (e.g., a junior researcher at an AI startup). Furthermore, we conduct a human-subject study to systematically evaluate the effectiveness of in-group persona agents in enhancing human-AI rapport. We compare our approach against two baseline conditions: a conventional agent without persona conditioning and an agent exhibiting minimal self-disclosure (e.g., "I've felt that too"). Results from post-task questionnaires assessing rapport and user experience indicate that the in-group persona agent significantly improves perceived rapport and personal relevance compared to the baselines, and also yields more positive user experience-most notably higher engagement.

cs.HC↗

arXiv2026

FlowFixer: Towards Detail-Preserving Subject-Driven Generation

We present FlowFixer, a refinement framework for subject-driven generation (SDG) that restores fine details lost during generation caused by changes in scale and perspective of a subject. FlowFixer proposes direct image-to-image translation from visual references, avoiding ambiguities in language prompts. To enable image-to-image training, we introduce a one-step denoising scheme to generate self-supervised training data, which automatically removes high-frequency details while preserving global structure, effectively simulating real-world SDG errors. We further propose a keypoint matching-based metric to properly assess fidelity in details beyond semantic similarities usually measured by CLIP or DINO. Experimental results demonstrate that FlowFixer outperforms state-of-the-art SDG methods in both qualitative and quantitative evaluations, setting a new benchmark for high-fidelity subject-driven generation.

cs.CV↗

arXiv2025

Negative-Guided Subject Fidelity Optimization for Zero-Shot Subject-Driven Generation

We present Subject Fidelity Optimization (SFO), a novel comparative learning framework for zero-shot subject-driven generation that enhances subject fidelity. Existing supervised fine-tuning methods, which rely only on positive targets and use the diffusion loss as in the pre-training stage, often fail to capture fine-grained subject details. To address this, SFO introduces additional synthetic negative targets and explicitly guides the model to favor positives over negatives through pairwise comparison. For negative targets, we propose Condition-Degradation Negative Sampling (CDNS), which automatically produces synthetic negatives tailored for subject-driven generation by introducing controlled degradations that emphasize subject fidelity and text alignment without expensive human annotations. Moreover, we reweight the diffusion timesteps to focus fine-tuning on intermediate steps where subject details emerge. Extensive experiments demonstrate that SFO with CDNS significantly outperforms recent strong baselines in terms of both subject fidelity and text alignment on a subject-driven generation benchmark. Project page: https://subjectfidelityoptimization.github.io/

cs.CV↗

arXiv2025

Know "No" Better: A Data-Driven Approach for Enhancing Negation Awareness in CLIP

While CLIP has significantly advanced multimodal understanding by bridging vision and language, the inability to grasp negation - such as failing to differentiate concepts like "parking" from "no parking" - poses substantial challenges. By analyzing the data used in the public CLIP model's pre-training, we posit this limitation stems from a lack of negation-inclusive data. To address this, we introduce data generation pipelines that employ a large language model (LLM) and a multimodal LLM to produce negation-inclusive captions. Fine-tuning CLIP with data generated from our pipelines, we develop NegationCLIP, which enhances negation awareness while preserving the generality. Moreover, to enable a comprehensive evaluation of negation understanding, we propose NegRefCOCOg-a benchmark tailored to test VLMs' ability to interpret negation across diverse expressions and positions within a sentence. Experiments on various CLIP architectures validate the effectiveness of our data generation pipelines in enhancing CLIP's ability to perceive negation accurately. Additionally, NegationCLIP's enhanced negation awareness has practical applications across various multimodal tasks, demonstrated by performance gains in text-to-image generation and referring image segmentation.

cs.CV↗

arXiv2025

Style-Friendly SNR Sampler for Style-Driven Generation

Recent text-to-image diffusion models generate high-quality images but struggle to learn new, personalized styles, which limits the creation of unique style templates. In style-driven generation, users typically supply reference images exemplifying the desired style, together with text prompts that specify desired stylistic attributes. Previous approaches popularly rely on fine-tuning, yet it often blindly utilizes objectives and noise level distributions from pre-training without adaptation. We discover that stylistic features predominantly emerge at higher noise levels, leading current fine-tuning methods to exhibit suboptimal style alignment. We propose the Style-friendly SNR sampler, which aggressively shifts the signal-to-noise ratio (SNR) distribution toward higher noise levels during fine-tuning to focus on noise levels where stylistic features emerge. This enhances models' ability to capture novel styles indicated by reference images and text prompts. We demonstrate improved generation of novel styles that cannot be adequately described solely with a text prompt, enabling the creation of new style templates for personalized content creation.

cs.CV↗

arXiv2024

Toward Interactive Regional Understanding in Vision-Large Language Models

Recent Vision-Language Pre-training (VLP) models have demonstrated significant advancements. Nevertheless, these models heavily rely on image-text pairs that capture only coarse and global information of an image, leading to a limitation in their regional understanding ability. In this work, we introduce \textbf{RegionVLM}, equipped with explicit regional modeling capabilities, allowing them to understand user-indicated image regions. To achieve this, we design a simple yet innovative architecture, requiring no modifications to the model architecture or objective function. Additionally, we leverage a dataset that contains a novel source of information, namely Localized Narratives, which has been overlooked in previous VLP research. Our experiments demonstrate that our single generalist model not only achieves an interactive dialogue system but also exhibits superior performance on various zero-shot region understanding tasks, without compromising its ability for global image understanding.

cs.CV↗

arXiv2023

Improving Visual Prompt Tuning for Self-supervised Vision Transformers

Visual Prompt Tuning (VPT) is an effective tuning method for adapting pretrained Vision Transformers (ViTs) to downstream tasks. It leverages extra learnable tokens, known as prompts, which steer the frozen pretrained ViTs. Although VPT has demonstrated its applicability with supervised vision transformers, it often underperforms with self-supervised ones. Through empirical observations, we deduce that the effectiveness of VPT hinges largely on the ViT blocks with which the prompt tokens interact. Specifically, VPT shows improved performance on image classification tasks for MAE and MoCo v3 when the prompt tokens are inserted into later blocks rather than the first block. These observations suggest that there exists an optimal location of blocks for the insertion of prompt tokens. Unfortunately, identifying the optimal blocks for prompts within each self-supervised ViT for diverse future scenarios is a costly process. To mitigate this problem, we propose a simple yet effective method that learns a gate for each ViT block to adjust its intervention into the prompt tokens. With our method, prompt tokens are selectively influenced by blocks that require steering for task adaptation. Our method outperforms VPT variants in FGVC and VTAB image classification and ADE20K semantic segmentation. The code is available at https://github.com/ryongithub/GatedPromptTuning.

cs.LG↗