Donghyun Kim

James Seale Smith, Leonid Karlinsky, Vyshnavi Gutta et al.

Computer vision models suffer from a phenomenon known as catastrophic forgetting when learning novel concepts from continuously shifting training data. Typical solutions for this continual learning problem require extensive rehearsal of previously seen data, which increases memory costs and may violate data privacy. Recently, the emergence of large-scale pre-trained vision transformer models has enabled prompting approaches as an alternative to data-rehearsal. These approaches rely on a key-query mechanism to generate prompts and have been found to be highly resistant to catastrophic forgetting in the well-established rehearsal-free continual learning setting. However, the key mechanism of these methods is not trained end-to-end with the task sequence. Our experiments show that this leads to a reduction in their plasticity, hence sacrificing new task accuracy, and inability to benefit from expanded parameter capacity. We instead propose to learn a set of prompt components which are assembled with input-conditioned weights to produce input-conditioned prompts, resulting in a novel attention-based end-to-end key-query scheme. Our experiments show that we outperform the current SOTA method DualPrompt on established benchmarks by as much as 4.5% in average final accuracy. We also outperform the state of art by as much as 4.4% accuracy on a continual learning benchmark which contains both class-incremental and domain-incremental task shifts, corresponding to many practical settings. Our code is available at https://github.com/GT-RIPL/CODA-Prompt

Donghyun Kim, Kaihong Wang, Kate Saenko et al.

While pose estimation is an important computer vision task, it requires expensive annotation and suffers from domain shift. In this paper, we investigate the problem of domain adaptive 2D pose estimation that transfers knowledge learned on a synthetic source domain to a target domain without supervision. While several domain adaptive pose estimation models have been proposed recently, they are not generic but only focus on either human pose or animal pose estimation, and thus their effectiveness is somewhat limited to specific scenarios. In this work, we propose a unified framework that generalizes well on various domain adaptive pose estimation problems. We propose to align representations using both input-level and output-level cues (pixels and pose labels, respectively), which facilitates the knowledge transfer from the source domain to the unlabeled target domain. Our experiments show that our method achieves state-of-the-art performance under various domain shifts. Our method outperforms existing baselines on human pose estimation by up to 4.5 percent points (pp), hand pose estimation by up to 7.4 pp, and animal pose estimation by up to 4.8 pp for dogs and 3.3 pp for sheep. These results suggest that our method is able to mitigate domain shift on diverse tasks and even unseen domains and objects (e.g., trained on horse and tested on dog). Our code will be publicly available at: https://github.com/VisionLearningGroup/UDA_PoseEstimation.

James Seale Smith, Paola Cascante-Bonilla, Assaf Arbelle et al.

Recently, large-scale pre-trained Vision-and-Language (VL) foundation models have demonstrated remarkable capabilities in many zero-shot downstream tasks, achieving competitive results for recognizing objects defined by as little as short text prompts. However, it has also been shown that VL models are still brittle in Structured VL Concept (SVLC) reasoning, such as the ability to recognize object attributes, states, and inter-object relations. This leads to reasoning mistakes, which need to be corrected as they occur by teaching VL models the missing SVLC skills; often this must be done using private data where the issue was found, which naturally leads to a data-free continual (no task-id) VL learning setting. In this work, we introduce the first Continual Data-Free Structured VL Concepts Learning (ConStruct-VL) benchmark and show it is challenging for many existing data-free CL strategies. We, therefore, propose a data-free method comprised of a new approach of Adversarial Pseudo-Replay (APR) which generates adversarial reminders of past tasks from past task models. To use this method efficiently, we also propose a continual parameter-efficient Layered-LoRA (LaLo) neural architecture allowing no-memory-cost access to all past models at train time. We show this approach outperforms all data-free methods by as much as ~7% while even matching some levels of experience-replay (prohibitive for applications where data-privacy must be preserved). Our code is publicly available at https://github.com/jamessealesmith/ConStruct-VL

Kuniaki Saito, Donghyun Kim, Piotr Teterwak et al.

Building object detectors that are robust to domain shifts is critical for real-world applications. Prior approaches fine-tune a pre-trained backbone and risk overfitting it to in-distribution (ID) data and distorting features useful for out-of-distribution (OOD) generalization. We propose to use Relative Gradient Norm (RGN) as a way to measure the vulnerability of a backbone to feature distortion, and show that high RGN is indeed correlated with lower OOD performance. Our analysis of RGN yields interesting findings: some backbones lose OOD robustness during fine-tuning, but others gain robustness because their architecture prevents the parameters from changing too much from the initial model. Given these findings, we present recipes to boost OOD robustness for both types of backbones. Specifically, we investigate regularization and architectural choices for minimizing gradient updates so as to prevent the tuned backbone from losing generalizable features. Our proposed techniques complement each other and show substantial improvements over baselines on diverse architectures and datasets. Code is available at https://github.com/VisionLearningGroup/mind_back.

Dong-Hee Kim, Reuben Tan, Donghyun Kim

Visual agents employ external visual tools within visual chains of thought to incorporate fine-grained evidence. While prior work has mainly studied these tools in visual search tasks, their role in more complex visual reasoning remains underexplored. In this paper, we move beyond simple visual search tasks to investigate more challenging tasks, including 3D spatial reasoning and medical visual question answering, where agents must integrate tool-acquired local evidence with the global context. We identify a {tool-use collapse phenomenon: models progressively stop using tools while still achieving higher task accuracy. Moreover, we observe a clear asymmetry: (i) completely eliminating tool use degrades performance, whereas (ii) incentivizing tool use yields only marginal gains despite substantially increasing usage. We find that vanilla training and tool-use encouragement both reduce rollout diversity, explaining why higher tool use does not yield stronger reasoning performance. Motivated by these findings, we add an entropy regularization term to encourage diverse rollout exploration, achieving the best performance despite gradually declining tool usage. % We further observe similar dynamics on medical VQA, suggesting that tool-use collapse is not limited to 3D spatial reasoning. Overall, our findings suggest a training-time view of tools as scaffolding, where broader exploration over language generation and visual tool invocation improves reasoning despite tool-use collapse. Project page: https://scaffolded-exploration.github.io

Howard Zhong, Samarth Mishra, Donghyun Kim et al.

Pre-training on massive video datasets has become essential to achieve high action recognition performance on smaller downstream datasets. However, most large-scale video datasets contain images of people and hence are accompanied with issues related to privacy, ethics, and data protection, often preventing them from being publicly shared for reproducible research. Existing work has attempted to alleviate these problems by blurring faces, downsampling videos, or training on synthetic data. On the other hand, analysis on the transferability of privacy-preserving pre-trained models to downstream tasks has been limited. In this work, we study this problem by first asking the question: can we pre-train models for human action recognition with data that does not include real humans? To this end, we present, for the first time, a benchmark that leverages real-world videos with humans removed and synthetic data containing virtual humans to pre-train a model. We then evaluate the transferability of the representation learned on this data to a diverse set of downstream action recognition benchmarks. Furthermore, we propose a novel pre-training strategy, called Privacy-Preserving MAE-Align, to effectively combine synthetic data and human-removed real data. Our approach outperforms previous baselines by up to 5% and closes the performance gap between human and no-human action recognition representations on downstream tasks, for both linear probing and fine-tuning. Our benchmark, code, and models are available at https://github.com/howardzh01/PPMA .

Donghyun Kim, Kaihong Wang, Stan Sclaroff et al.

Deep models must learn robust and transferable representations in order to perform well on new domains. While domain transfer methods (e.g., domain adaptation, domain generalization) have been proposed to learn transferable representations across domains, they are typically applied to ResNet backbones pre-trained on ImageNet. Thus, existing works pay little attention to the effects of pre-training on domain transfer tasks. In this paper, we provide a broad study and in-depth analysis of pre-training for domain adaptation and generalization, namely: network architectures, size, pre-training loss, and datasets. We observe that simply using a state-of-the-art backbone outperforms existing state-of-the-art domain adaptation baselines and set new baselines on Office-Home and DomainNet improving by 10.7\% and 5.5\%. We hope that this work can provide more insights for future domain transfer research.

Thao Nguyen, Krishna Kumar Singh, Donghyun Kim et al.

We study the personal camera roll visual question answering setting. In this setting, a conversational AI assistant can access a user's personal camera roll and retrieve relevant photos to answer queries, ranging from simple factual questions (e.g., ``Name of the food I tried yesterday?'') to more open-ended ones (e.g., ``Recommend some dishes I have never eaten before''). Given the vast nature of the personal camera roll (i.e., multiple years, hundreds to thousands of photos), a successful AI assistant needs to understand a long-horizon, highly personalized visual content stream in order to navigate and locate the correct and/or relevant information. To support this, we collect and manually annotate questions that mimic real-world usage. The final dataset, camroll, contains 50 users, 31,476 images, and 2,500 QA pairs. We further design camroll-agent, a conversational AI agent equipped with hierarchical memory and a minimal set of tools for efficient navigation over large, personalized visual memory. Experimental results show that camroll-agent outperforms numerous baselines and methods for long-context understanding AI agents system. Together, the camroll dataset and camroll-agent highlight the gap in AI agents' long-context reasoning: personalized visual memory requires different approaches from standard long-context textual memory, especially when consistency, visual details, and user-specific context are present.

Daehee Kim, Yoonsik Kim, DongHyun Kim et al.

Inspired by the great success of language model (LM)-based pre-training, recent studies in visual document understanding have explored LM-based pre-training methods for modeling text within document images. Among them, pre-training that reads all text from an image has shown promise, but often exhibits instability and even fails when applied to broader domains, such as those involving both visual documents and scene text images. This is a substantial limitation for real-world scenarios, where the processing of text image inputs in diverse domains is essential. In this paper, we investigate effective pre-training tasks in the broader domains and also propose a novel pre-training method called SCOB that leverages character-wise supervised contrastive learning with online text rendering to effectively pre-train document and scene text domains by bridging the domain gap. Moreover, SCOB enables weakly supervised learning, significantly reducing annotation costs. Extensive benchmarks demonstrate that SCOB generally improves vanilla pre-training methods and achieves comparable performance to state-of-the-art methods. Our findings suggest that SCOB can be served generally and effectively for read-type pre-training methods. The code will be available at https://github.com/naver-ai/scob.

Sein Kim, Namkyeong Lee, Donghyun Kim et al.

User modeling, which learns to represent users into a low-dimensional representation space based on their past behaviors, got a surge of interest from the industry for providing personalized services to users. Previous efforts in user modeling mainly focus on learning a task-specific user representation that is designed for a single task. However, since learning task-specific user representations for every task is infeasible, recent studies introduce the concept of universal user representation, which is a more generalized representation of a user that is relevant to a variety of tasks. Despite their effectiveness, existing approaches for learning universal user representations are impractical in real-world applications due to the data requirement, catastrophic forgetting and the limited learning capability for continually added tasks. In this paper, we propose a novel continual user representation learning method, called TERACON, whose learning capability is not limited as the number of learned tasks increases while capturing the relationship between the tasks. The main idea is to introduce an embedding for each task, i.e., task embedding, which is utilized to generate task-specific soft masks that not only allow the entire model parameters to be updated until the end of training sequence, but also facilitate the relationship between the tasks to be captured. Moreover, we introduce a novel knowledge retention module with pseudo-labeling strategy that successfully alleviates the long-standing problem of continual learning, i.e., catastrophic forgetting. Extensive experiments on public and proprietary real-world datasets demonstrate the superiority and practicality of TERACON. Our code is available at https://github.com/Sein-Kim/TERACON.

Sivan Doveh, Assaf Arbelle, Sivan Harary et al.

Vision and Language (VL) models have demonstrated remarkable zero-shot performance in a variety of tasks. However, some aspects of complex language understanding still remain a challenge. We introduce the collective notion of Structured Vision&Language Concepts (SVLC) which includes object attributes, relations, and states which are present in the text and visible in the image. Recent studies have shown that even the best VL models struggle with SVLC. A possible way of fixing this issue is by collecting dedicated datasets for teaching each SVLC type, yet this might be expensive and time-consuming. Instead, we propose a more elegant data-driven approach for enhancing VL models' understanding of SVLCs that makes more effective use of existing VL pre-training datasets and does not require any additional data. While automatic understanding of image structure still remains largely unsolved, language structure is much better modeled and understood, allowing for its effective utilization in teaching VL models. In this paper, we propose various techniques based on language structure understanding that can be used to manipulate the textual part of off-the-shelf paired VL datasets. VL models trained with the updated data exhibit a significant improvement of up to 15% in their SVLC understanding with only a mild degradation in their zero-shot capabilities both when training from scratch or fine-tuning a pre-trained model.

Paola Cascante-Bonilla, Khaled Shehada, James Seale Smith et al.

Large-scale pre-trained Vision & Language (VL) models have shown remarkable performance in many applications, enabling replacing a fixed set of supported classes with zero-shot open vocabulary reasoning over (almost arbitrary) natural language prompts. However, recent works have uncovered a fundamental weakness of these models. For example, their difficulty to understand Visual Language Concepts (VLC) that go 'beyond nouns' such as the meaning of non-object words (e.g., attributes, actions, relations, states, etc.), or difficulty in performing compositional reasoning such as understanding the significance of the order of the words in a sentence. In this work, we investigate to which extent purely synthetic data could be leveraged to teach these models to overcome such shortcomings without compromising their zero-shot capabilities. We contribute Synthetic Visual Concepts (SyViC) - a million-scale synthetic dataset and data generation codebase allowing to generate additional suitable data to improve VLC understanding and compositional reasoning of VL models. Additionally, we propose a general VL finetuning strategy for effectively leveraging SyViC towards achieving these improvements. Our extensive experiments and ablations on VL-Checklist, Winoground, and ARO benchmarks demonstrate that it is possible to adapt strong pre-trained VL models with synthetic data significantly enhancing their VLC understanding (e.g. by 9.9% on ARO and 4.3% on VL-Checklist) with under 1% drop in their zero-shot accuracy.

Jianrui Zhang, Hyun Jung Lee, Sukanta Ganguly et al.

Multimodal retrieval relies heavily on single-vector retrievers, which compress rich, sequential token sequences into one single global representation. While efficient, they discard fine-grained, local evidence critical for dense retrieval tasks. Multi-vector approaches were introduced as a solution, but they strictly require training and many ignore the necessity of a globally summarizing representation. To address this, we introduce SMART, a framework that unlocks the latent multi-vector capabilities of standard single-vector models. We first demonstrate that standard contrastive training on the pooled embedding implicitly shapes the retrieval geometry of preceding hidden states via gradient flow. By applying direct late-interaction over these frozen hidden states during inference, SMART acts as a plug-and-play upgrade that consistently improves performance across diverse modalities, improving even the state-of-the-art models further on MMEB-V2. We also reveal SMART's superior performance, as simple lightweight post-training not only saves time and compute, but also brings forth further improvement on Visual Document retrieval, allowing a single-vector model to outperform SoTA multi-vector counterparts. Ultimately, SMART offers both a highly efficient inference enhancement and a powerful finetuning technique for multimodal retrieval. We open source our code and weights at https://github.com/HanSolo9682/SMART.

Kaihong Wang, Donghyun Kim, Rogerio Feris et al.

While transformers have greatly boosted performance in semantic segmentation, domain adaptive transformers are not yet well explored. We identify that the domain gap can cause discrepancies in self-attention. Due to this gap, the transformer attends to spurious regions or pixels, which deteriorates accuracy on the target domain. We propose to perform adaptation on attention maps with cross-domain attention layers that share features between the source and the target domains. Specifically, we impose consistency between predictions from cross-domain attention and self-attention modules to encourage similar distribution in the attention and output of the model across domains, i.e., attention-level and output-level alignment. We also enforce consistency in attention maps between different augmented views to further strengthen the attention-based alignment. Combining these two components, our method mitigates the discrepancy in attention maps across domains and further boosts the performance of the transformer under unsupervised domain adaptation settings. Our model outperforms the existing state-of-the-art baseline model on three widely used benchmarks, including GTAV-to-Cityscapes by 1.3 percent point (pp), Synthia-to-Cityscapes by 0.6 pp, and Cityscapes-to-ACDC by 1.1 pp, on average. Additionally, we verify the effectiveness and generalizability of our method through extensive experiments. Our code will be publicly available.

Zhongyang Zhang, Kaidong Chai, Haowen Yu et al.

As a beloved sport worldwide, dancing is getting integrated into traditional and virtual reality-based gaming platforms nowadays. It opens up new opportunities in the technology-mediated dancing space. These platforms primarily rely on passive and continuous human pose estimation as an input capture mechanism. Existing solutions are mainly based on RGB or RGB-Depth cameras for dance games. The former suffers in low-lighting conditions due to the motion blur and low sensitivity, while the latter is too power-hungry, has a low frame rate, and has limited working distance. With ultra-low latency, energy efficiency, and wide dynamic range characteristics, the event camera is a promising solution to overcome these shortcomings. We propose YeLan, an event camera-based 3-dimensional high-frequency human pose estimation(HPE) system that survives low-lighting conditions and dynamic backgrounds. We collected the world's first event camera dance dataset and developed a fully customizable motion-to-event physics-aware simulator. YeLan outperforms the baseline models in these challenging conditions and demonstrated robustness against different types of clothing, background motion, viewing angle, occlusion, and lighting fluctuations.

Dina Bashkirova, Samarth Mishra, Diala Lteif et al.

Label-efficient and reliable semantic segmentation is essential for many real-life applications, especially for industrial settings with high visual diversity, such as waste sorting. In industrial waste sorting, one of the biggest challenges is the extreme diversity of the input stream depending on factors like the location of the sorting facility, the equipment available in the facility, and the time of year, all of which significantly impact the composition and visual appearance of the waste stream. These changes in the data are called ``visual domains'', and label-efficient adaptation of models to such domains is needed for successful semantic segmentation of industrial waste. To test the abilities of computer vision models on this task, we present the VisDA 2022 Challenge on Domain Adaptation for Industrial Waste Sorting. Our challenge incorporates a fully-annotated waste sorting dataset, ZeroWaste, collected from two real material recovery facilities in different locations and seasons, as well as a novel procedurally generated synthetic waste sorting dataset, SynthWaste. In this competition, we aim to answer two questions: 1) can we leverage domain adaptation techniques to minimize the domain gap? and 2) can synthetic data augmentation improve performance on this task and help adapt to changing data distributions? The results of the competition show that industrial waste detection poses a real domain adaptation problem, that domain generalization techniques such as augmentations, ensembling, etc., improve the overall performance on the unlabeled target domain examples, and that leveraging synthetic data effectively remains an open problem. See https://ai.bu.edu/visda-2022/

Hochul Hwang, Tim Xia, Ibrahima Keita et al.

A robot guide dog has compelling advantages over animal guide dogs for its cost-effectiveness, potential for mass production, and low maintenance burden. However, despite the long history of guide dog robot research, previous studies were conducted with little or no consideration of how the guide dog handler and the guide dog work as a team for navigation. To develop a robotic guiding system that is genuinely beneficial to blind or visually impaired individuals, we performed qualitative research, including interviews with guide dog handlers and trainers and first-hand blindfold walking experiences with various guide dogs. Grounded on the facts learned from vivid experience and interviews, we build a collaborative indoor navigation scheme for a guide dog robot that includes preferred features such as speed and directional control. For collaborative navigation, we propose a semantic-aware local path planner that enables safe and efficient guiding work by utilizing semantic information about the environment and considering the handler's position and directional cues to determine the collision-free path. We evaluate our integrated robotic system by testing guide blindfold walking in indoor settings and demonstrate guide dog-like navigation behavior by avoiding obstacles at typical gait speed ($0.7 \mathrm{m/s}$).

Donghyun Kim, Teakgyu Hong, Moonbin Yim et al.

In recent years, research on visual document understanding (VDU) has grown significantly, with a particular emphasis on the development of self-supervised learning methods. However, one of the significant challenges faced in this field is the limited availability of publicly accessible visual corpora or extensive collections of images with detailed text annotations, particularly for non-Latin or resource-scarce languages. To address this challenge, we propose Web-based Visual Corpus Builder (Webvicob), a dataset generator engine capable of constructing large-scale, multilingual visual corpora from raw Wikipedia HTML dumps. Our experiments demonstrate that the data generated by Webvicob can be used to train robust VDU models that perform well on various downstream tasks, such as DocVQA and post-OCR parsing. Furthermore, when using a dataset of 1 million images generated by Webvicob, we observed an improvement of over 13% on the DocVQA Task 3 compared to a dataset of 11 million images from the IIT-CDIP. The implementation of our engine is publicly available on https://github.com/clovaai/webvicob

Quanfu Fan, Donghyun Kim, Chun-Fu et al.

In this paper, we provide a deep analysis of temporal modeling for action recognition, an important but underexplored problem in the literature. We first propose a new approach to quantify the temporal relationships between frames captured by CNN-based action models based on layer-wise relevance propagation. We then conduct comprehensive experiments and in-depth analysis to provide a better understanding of how temporal modeling is affected by various factors such as dataset, network architecture, and input frames. With this, we further study some important questions for action recognition that lead to interesting findings. Our analysis shows that there is no strong correlation between temporal relevance and model performance; and action models tend to capture local temporal information, but less long-range dependencies. Our codes and models will be publicly available.

Jongwoo Park, Kumara Kahatapitiya, Donghyun Kim et al.

Vision Transformers (ViTs) have recently become the state-of-the-art across many computer vision tasks. In contrast to convolutional networks (CNNs), ViTs enable global information sharing even within shallow layers of a network, i.e., among high-resolution features. However, this perk was later overlooked with the success of pyramid architectures such as Swin Transformer, which show better performance-complexity trade-offs. In this paper, we present a simple and efficient add-on component (termed GrafT) that considers global dependencies and multi-scale information throughout the network, in both high- and low-resolution features alike. It has the flexibility of branching out at arbitrary depths and shares most of the parameters and computations of the backbone. GrafT shows consistent gains over various well-known models which includes both hybrid and pure Transformer types, both homogeneous and pyramid structures, and various self-attention methods. In particular, it largely benefits mobile-size models by providing high-level semantics. On the ImageNet-1k dataset, GrafT delivers +3.9%, +1.4%, and +1.9% top-1 accuracy improvement to DeiT-T, Swin-T, and MobileViT-XXS, respectively. Our code and models will be made available.

Donghyun Kim, Junhyeok Ahn, Orion Campbell et al.

We design, build, and thoroughly test a new type of actuator dubbed viscoelastic liquid cooled actuator (VLCA) for robotic applications. VLCAs excel in the following five critical axes of performance: energy efficiency, torque density, impact resistence, joint position and force controllability. We first study the design objectives and choices of the VLCA to enhance the performance on the needed criteria. We follow by an investigation on viscoelastic materials in terms of their damping, viscous and hysteresis properties as well as parameters related to the long- term performance. As part of the actuator design, we configure a disturbance observer to provide high-fidelity force control to enable a wide range of impedance control capabilities. We proceed to design a robotic system capable to lift payloads of 32.5 kg, which is three times larger than its own weight. In addition, we experiment with Cartesian trajectory control up to 2 Hz with a vertical range of motion of 32 cm while carrying a payload of 10 kg. Finally, we perform experiments on impedance control and mechanical robustness by studying the response of the robotics testbed to hammering impacts and external force interactions.

Sungyeon Kim, Boseung Jeong, Donghyun Kim et al.

Large-scale image-text pre-trained models enable zero-shot classification and provide consistent accuracy across various data distributions. Nonetheless, optimizing these models in downstream tasks typically requires fine-tuning, which reduces generalization to out-of-distribution (OOD) data and demands extensive computational resources. We introduce Robust Adapter (R-Adapter), a novel method for fine-tuning zero-shot models to downstream tasks while simultaneously addressing both these issues. Our method integrates lightweight modules into the pre-trained model and employs novel self-ensemble techniques to boost OOD robustness and reduce storage expenses substantially. Furthermore, we propose MPM-NCE loss designed for fine-tuning on vision-language downstream tasks. It ensures precise alignment of multiple image-text pairs and discriminative feature learning. By extending the benchmark for robust fine-tuning beyond classification to include diverse tasks such as cross-modal retrieval and open vocabulary segmentation, we demonstrate the broad applicability of R-Adapter. Our extensive experiments demonstrate that R-Adapter achieves state-of-the-art performance across a diverse set of tasks, tuning only 13% of the parameters of the CLIP encoders.

Yongwoo Kim, Sojung An, Yunjin Park et al.

Multimodal Large Language Models (MLLMs) exacerbate safety risks by introducing vulnerabilities across multiple modalities, such as language and vision. Current MLLM safety evaluation tools, however, suffer from major limitations: 1) English-centric dataset construction, and 2) a focus on generic risks that are not tied to local cultural contexts. This paper introduces KSAFE-MM, a benchmark for Korean multimodal safety evaluation that covers both general safety risks and culture-specific vulnerabilities. KSAFE-MM consists of two parts, KSAFE-MM-G and KSAFE-MM-C. KSAFE-MM-G evaluates globally shared risks in Korean contexts through linguistic contextualization, which transforms generic safety queries into contextually grounded multimodal samples. KSAFE-MM-C targets culture-dependent MLLM safety vulnerabilities using localized visual queries derived from real-world contexts. It pairs these visual queries with jailbreak-style textual queries to cover multimodal safety risks involving cultural visual cues and malicious textual intent. Together, these components provide a general-to-local construction pipeline for evaluating both globally shared safety risks and culture-specific vulnerabilities. We evaluate 12 state-of-the-art MLLMs on KSAFE-MM and reveal that models exhibit greater vulnerability to culturally grounded attacks than to generic ones. Notably, jailbreaking strategies substantially amplify attack success rates, with ProgramExecution yielding up to 74.2% ASR compared to 13.4% for standard queries. Furthermore, we identify a systematic trade-off between safety and over-refusal, where models achieving low ASR tend to exhibit excessive refusal behavior on benign queries. These findings highlight the urgent need for culturally grounded safety evaluation beyond English-centric benchmarks.

Kibum Kim, Kanghoon Yoon, Jaehyeong Jeon et al.

Weakly-Supervised Scene Graph Generation (WSSGG) research has recently emerged as an alternative to the fully-supervised approach that heavily relies on costly annotations. In this regard, studies on WSSGG have utilized image captions to obtain unlocalized triplets while primarily focusing on grounding the unlocalized triplets over image regions. However, they have overlooked the two issues involved in the triplet formation process from the captions: 1) Semantic over-simplification issue arises when extracting triplets from captions, where fine-grained predicates in captions are undesirably converted into coarse-grained predicates, resulting in a long-tailed predicate distribution, and 2) Low-density scene graph issue arises when aligning the triplets in the caption with entity/predicate classes of interest, where many triplets are discarded and not used in training, leading to insufficient supervision. To tackle the two issues, we propose a new approach, i.e., Large Language Model for weakly-supervised SGG (LLM4SGG), where we mitigate the two issues by leveraging the LLM's in-depth understanding of language and reasoning ability during the extraction of triplets from captions and alignment of entity/predicate classes with target data. To further engage the LLM in these processes, we adopt the idea of Chain-of-Thought and the in-context few-shot learning strategy. To validate the effectiveness of LLM4SGG, we conduct extensive experiments on Visual Genome and GQA datasets, showing significant improvements in both Recall@K and mean Recall@K compared to the state-of-the-art WSSGG methods. A further appeal is that LLM4SGG is data-efficient, enabling effective model training with a small amount of training images.

Sungyeon Kim, Donghyun Kim, Suha Kwak

A common practice in metric learning is to train and test an embedding model for each dataset. This dataset-specific approach fails to simulate real-world scenarios that involve multiple heterogeneous distributions of data. In this regard, we explore a new metric learning paradigm, called Unified Metric Learning (UML), which learns a unified distance metric capable of capturing relations across multiple data distributions. UML presents new challenges, such as imbalanced data distribution and bias towards dominant distributions. These issues cause standard metric learning methods to fail in learning a unified metric. To address these challenges, we propose Parameter-efficient Unified Metric leArning (PUMA), which consists of a pre-trained frozen model and two additional modules, stochastic adapter and prompt pool. These modules enable to capture dataset-specific knowledge while avoiding bias towards dominant distributions. Additionally, we compile a new unified metric learning benchmark with a total of 8 different datasets. PUMA outperforms the state-of-the-art dataset-specific models while using about 69 times fewer trainable parameters.

Wongyu Kim, Youbin Ahn, Donghyun Kim et al.

Each utterance in multi-turn empathetic dialogues has features such as emotion, keywords, and utterance-level meaning. Feature transitions between utterances occur naturally. However, existing approaches fail to perceive the transitions because they extract features for the context at the coarse-grained level. To solve the above issue, we propose a novel approach of recognizing feature transitions between utterances, which helps understand the dialogue flow and better grasp the features of utterance that needs attention. Also, we introduce a response generation strategy to help focus on emotion and keywords related to appropriate features when generating responses. Experimental results show that our approach outperforms baselines and especially, achieves significant improvements on multi-turn dialogues.

Sein Kim, Hongseok Kang, Seungyoon Choi et al.

Collaborative filtering recommender systems (CF-RecSys) have shown successive results in enhancing the user experience on social media and e-commerce platforms. However, as CF-RecSys struggles under cold scenarios with sparse user-item interactions, recent strategies have focused on leveraging modality information of user/items (e.g., text or images) based on pre-trained modality encoders and Large Language Models (LLMs). Despite their effectiveness under cold scenarios, we observe that they underperform simple traditional collaborative filtering models under warm scenarios due to the lack of collaborative knowledge. In this work, we propose an efficient All-round LLM-based Recommender system, called A-LLMRec, that excels not only in the cold scenario but also in the warm scenario. Our main idea is to enable an LLM to directly leverage the collaborative knowledge contained in a pre-trained state-of-the-art CF-RecSys so that the emergent ability of the LLM as well as the high-quality user/item embeddings that are already trained by the state-of-the-art CF-RecSys can be jointly exploited. This approach yields two advantages: (1) model-agnostic, allowing for integration with various existing CF-RecSys, and (2) efficiency, eliminating the extensive fine-tuning typically required for LLM-based recommenders. Our extensive experiments on various real-world datasets demonstrate the superiority of A-LLMRec in various scenarios, including cold/warm, few-shot, cold user, and cross-domain scenarios. Beyond the recommendation task, we also show the potential of A-LLMRec in generating natural language outputs based on the understanding of the collaborative knowledge by performing a favorite genre prediction task. Our code is available at https://github.com/ghdtjr/A-LLMRec .

Donghyun Kim, Chanyoung Kim, Eunseo Jeong et al.

Recent advances in monocular 3D human pose estimation enable accurate body tracking from video. However, translating these kinematic estimates into physical quantities, such as joint torques, remains challenging due to noise amplification through inverse dynamics. In this work, we provide a systematic analysis of how pose estimation noise propagates through the inverse dynamics pipeline. We present three key findings: (1) pose noise is amplified by approximately 1,000x when computing joint torques via numerical differentiation, (2) proximal joints (spine, hips) are up to 10x more sensitive to noise than distal joints (wrists, hands), and (3) low-pass filtering before differentiation substantially reduces this amplification. To enable this analysis, we develop SMPL-Dynamics, a fully differentiable inverse dynamics module for the SMPL body model that requires no external physics simulators. Our module supports end-to-end gradient computation, and we demonstrate this through differentiable pose refinement, which reduces torque error by 93% with negligible change in pose.

Minjun Park, Donghyun Kim, Hyeonjong Ju et al.

We are entering an era in which individuals and organizations increasingly deploy dedicated AI agents that interact and collaborate with other agents. However, the dynamics of multi-agent collaboration under privacy constraints remain poorly understood. In this work, we present $PAC\text{-}Bench$, a benchmark for systematic evaluation of multi-agent collaboration under privacy constraints. Experiments on $PAC\text{-}Bench$ show that privacy constraints substantially degrade collaboration performance and make outcomes depend more on the initiating agent than the partner. Further analysis reveals that this degradation is driven by recurring coordination breakdowns, including early-stage privacy violations, overly conservative abstraction, and privacy-induced hallucinations. Together, our findings identify privacy-aware multi-agent collaboration as a distinct and unresolved challenge that requires new coordination mechanisms beyond existing agent capabilities.

Donghyun Kim, Byeongho Heo, Dongyoon Han

This paper revives Densely Connected Convolutional Networks (DenseNets) and reveals the underrated effectiveness over predominant ResNet-style architectures. We believe DenseNets' potential was overlooked due to untouched training methods and traditional design elements not fully revealing their capabilities. Our pilot study shows dense connections through concatenation are strong, demonstrating that DenseNets can be revitalized to compete with modern architectures. We methodically refine suboptimal components - architectural adjustments, block redesign, and improved training recipes towards widening DenseNets and boosting memory efficiency while keeping concatenation shortcuts. Our models, employing simple architectural elements, ultimately surpass Swin Transformer, ConvNeXt, and DeiT-III - key architectures in the residual learning lineage. Furthermore, our models exhibit near state-of-the-art performance on ImageNet-1K, competing with the very recent models and downstream tasks, ADE20k semantic segmentation, and COCO object detection/instance segmentation. Finally, we provide empirical analyses that uncover the merits of the concatenation over additive shortcuts, steering a renewed preference towards DenseNet-style designs. Our code is available at https://github.com/naver-ai/rdnet.

Kwanyong Park, Kuniaki Saito, Donghyun Kim

Vision-language (VL) models often exhibit a limited understanding of complex expressions of visual objects (e.g., attributes, shapes, and their relations), given complex and diverse language queries. Traditional approaches attempt to improve VL models using hard negative synthetic text, but their effectiveness is limited. In this paper, we harness the exceptional compositional understanding capabilities of generative foundational models. We introduce a novel method for structured synthetic data generation aimed at enhancing the compositional understanding of VL models in language-based object detection. Our framework generates densely paired positive and negative triplets (image, text descriptions, and bounding boxes) in both image and text domains. By leveraging these synthetic triplets, we transform 'weaker' VL models into 'stronger' models in terms of compositional understanding, a process we call "Weak-to-Strong Compositional Learning" (WSCL). To achieve this, we propose a new compositional contrastive learning formulation that discovers semantics and structures in complex descriptions from synthetic triplets. As a result, VL models trained with our synthetic data generation exhibit a significant performance boost in the Omnilabel benchmark by up to +5AP and the D3 benchmark by +6.9AP upon existing baselines.

Sehun Jung, HyunJee Song, Dong-Hee Kim et al.

Recent advances in robot manipulation increasingly leverage Vision-Language Models (VLMs) for high-level reasoning, such as decomposing task instructions into sequential action plans expressed in natural language that guide downstream low-level motor execution. However, current benchmarks do not assess whether these plans are spatially executable, particularly in specifying the exact spatial locations where the robot should interact to execute the plan, limiting evaluation of real-world manipulation capability. To bridge this gap, we define a novel task of grounded planning and introduce GroundedPlanBench, a newly curated benchmark for spatially grounded long-horizon action planning in the wild. GroundedPlanBench jointly evaluates hierarchical sub-action planning and spatial action grounding (where to act), enabling systematic assessment of whether generated sub-actions are spatially executable for robot manipulation. We further introduce Video-to-Spatially Grounded Planning (V2GP), an automated data generation framework that leverages real-world robot video demonstrations to improve spatially grounded long-horizon planning. Our evaluations reveal that spatially grounded long-horizon planning remains a major bottleneck for current VLMs. Our results demonstrate that V2GP provides a promising approach for improving both action planning and spatial grounding performance, validated on our benchmark as well as through real-world robot manipulation experiments, advancing progress toward spatially actionable planning.

Sein Kim, Hongseok Kang, Kibum Kim et al.

Large Language Models (LLMs) have recently emerged as promising tools for recommendation thanks to their advanced textual understanding ability and context-awareness. Despite the current practice of training and evaluating LLM-based recommendation (LLM4Rec) models under a sequential recommendation scenario, we found that whether these models understand the sequential information inherent in users' item interaction sequences has been largely overlooked. In this paper, we first demonstrate through a series of experiments that existing LLM4Rec models do not fully capture sequential information both during training and inference. Then, we propose a simple yet effective LLM-based sequential recommender, called LLM-SRec, a method that enhances the integration of sequential information into LLMs by distilling the user representations extracted from a pre-trained CF-SRec model into LLMs. Our extensive experiments show that LLM-SRec enhances LLMs' ability to understand users' item interaction sequences, ultimately leading to improved recommendation performance. Furthermore, unlike existing LLM4Rec models that require fine-tuning of LLMs, LLM-SRec achieves state-of-the-art performance by training only a few lightweight MLPs, highlighting its practicality in real-world applications. Our code is available at https://github.com/Sein-Kim/LLM-SRec.

Hochul Hwang, Krisha Adhikari, Satya Shodhaka et al.

Robotic mobility aids for blind and low-vision (BLV) individuals rely heavily on deep learning-based vision models specialized for various navigational tasks. However, the performance of these models is often constrained by the availability and diversity of real-world datasets, which are challenging to collect in sufficient quantities for different tasks. In this study, we investigate the effectiveness of synthetic data, generated using Unreal Engine 4, for training robust vision models for this safety-critical application. Our findings demonstrate that synthetic data can enhance model performance across multiple tasks, showcasing both its potential and its limitations when compared to real-world data. We offer valuable insights into optimizing synthetic data generation for developing robotic mobility aids. Additionally, we publicly release our generated synthetic dataset to support ongoing research in assistive technologies for BLV individuals, available at https://hchlhwang.github.io/SToP.

Donghyun Kim, Seil Kang, Seong Jae Hwang

Image dehazing, addressing atmospheric interference like fog and haze, remains a pervasive challenge crucial for robust vision applications such as surveillance and remote sensing under adverse visibility. While various methodologies have evolved from early works predicting transmission matrix and atmospheric light features to deep learning and dehazing networks, they innately prioritize dehazing quality metrics, neglecting the need for real-time applicability in time-sensitive domains like autonomous driving. This work introduces FALCON (Frequency Adjoint Link with CONtinuous density mask), a single-image dehazing system achieving state-of-the-art performance on both quality and speed. Particularly, we develop a novel bottleneck module, namely, Frequency Adjoint Link, operating in the frequency space to globally expand the receptive field with minimal growth in network size. Further, we leverage the underlying haze distribution based on the atmospheric scattering model via a Continuous Density Mask (CDM) which serves as a continuous-valued mask input prior and a differentiable auxiliary loss. Comprehensive experiments involving multiple state-of-the-art methods and ablation analysis demonstrate FALCON's exceptional performance in both dehazing quality and speed (i.e., >$180 frames-per-second), quantified by metrics such as FPS, PSNR, and SSIM.

Shangqun Yu, Matthew En, Daniel Wu et al.

Opening heavy, self closing doors, especially those that require pulling remains a long standing challenge in robotics. Humans naturally employ both arms in a dexterous manner, rotating the handle, widening the gap, holding the door, switching arms when needed, and moving through while maintaining clearance. To replicate such behaviors, a robot must perform a long sequence of motions spanning multiple stages and interactions with different parts of the door. Traditional approaches rely on state machines that transition between manually defined stages (e.g., pulling after the knob is rotated, passing after the gap is sufficiently wide). While intuitive, these methods lack robustness, as hand crafted trajectories fail to generalize to the diversity of real world conditions without extensive engineering effort. Recent advances in imitation learning offer a scalable alternative, yet no existing visual action model has demonstrated simultaneous coordination of a nonholonomic base and dual arms for the complete door opening and passing task. In this paper, we tackle this complex, highly constrained problem using a diffusion based visuomotor control policy. Our results demonstrate that a single end to end policy can be learned to execute long horizon tasks requiring tight coordination between manipulation and locomotion. The resulting policy not only achieves a high success rate in opening and traversing damped pull doors but also demonstrates strong robustness to external disturbances capabilities that are difficult to realize with traditional methods.

Chanyoung Kim, Donghyun Kim, Dong-Hyun Sim et al.

Graph convolutional networks (GCNs) are widely used for 3D hand pose estimation, where the hand skeleton is encoded as a fixed adjacency graph. We revisit whether this is the most effective way to incorporate hand topology in 2D-to-3D lifting. In this paper, we perform controlled, parameter-matched ablations on the FPHA benchmark and show that standard multi-head self-attention consistently outperforms GCN baselines. Even when the GCN is strengthened with multi-hop adjacency and matched parameter count, self-attention reduces MPJPE from 12.36 mm to 10.09 mm. A skeleton-constrained graph attention network recovers most of this gap, indicating that input-dependent aggregation is a major source of improvement, while fully connected attention yields additional gains. We further show that hand topology is most effective when introduced as a soft structural prior through graph-distance positional encoding, rather than as a hard adjacency constraint. These results suggest that, for hand pose lifting, adaptive spatial attention is a more effective inductive bias than fixed graph convolution.

Dong-Hee Kim, Hyunjee Song, Donghyun Kim

Despite the advances in Referring Expression Segmentation (RES) benchmarks, their evaluation protocols remain constrained, primarily focusing on either single targets with short queries (containing minimal attributes) or multiple targets from distinctly different queries on a single domain. This limitation significantly hinders the assessment of more complex reasoning capabilities in RES models. We introduce WildRES, a novel benchmark that incorporates long queries with diverse attributes and non-distinctive queries for multiple targets. This benchmark spans diverse application domains, including autonomous driving environments and robotic manipulation scenarios, thus enabling more rigorous evaluation of complex reasoning capabilities in real-world settings. Our analysis reveals that current RES models demonstrate substantial performance deterioration when evaluated on WildRES. To address this challenge, we introduce SynRES, an automated pipeline generating densely paired compositional synthetic training data through three innovations: (1) a dense caption-driven synthesis for attribute-rich image-mask-expression triplets, (2) reliable semantic alignment mechanisms rectifying caption-pseudo mask inconsistencies via Image-Text Aligned Grouping, and (3) domain-aware augmentations incorporating mosaic composition and superclass replacement to emphasize generalization ability and distinguishing attributes over object categories. Experimental results demonstrate that models trained with SynRES achieve state-of-the-art performance, improving gIoU by 2.0% on WildRES-ID and 3.8% on WildRES-DS. Code and datasets are available at https://github.com/UTLLab/SynRES.

Donghyun Kim, Jaehyoung Park

Autonomous driving scenes range from empty highways to dense intersections with dozens of interacting road users, yet current 3D detection models apply a fixed computation budget to every frame, wasting resources on simple scenes while lacking capacity for complex ones. Existing approaches compound this problem: Transformer-based interaction models scale quadratically with the number of detected objects, and frame-by-frame processing causes the system to immediately forget objects the moment they become occluded. We propose Enhanced HOPE, an adaptive perception architecture that measures the geometric complexity of each incoming LiDAR frame using an unsupervised statistical estimator and routes it through a shallow or deep processing path accordingly, requiring no manual scene labels. To keep interaction modeling efficient, we replace quadratic pairwise attention with a linear-time subspace-based network that groups nearby objects into clusters and processes them jointly. The computational savings from these two mechanisms free up resources for a persistent temporal memory module that retains previously detected objects and traffic rules across frames, enabling the system to recall occluded objects seconds after they disappear from view. On the nuScenes and CARLA benchmarks, Enhanced HOPE reduces latency by 38% on simple scenes with no accuracy loss, improves mean Average Precision by 2.7 points on rare long-tail scenarios, and tracks objects through occlusions lasting over 5 seconds, where all tested baselines fail.

Hochul Hwang, Soowan Yang, Jahir Sadik Monon et al.

While commendable progress has been made in user-centric research on mobile assistive systems for blind and low-vision (BLV) individuals, references that directly inform robot navigation design remain rare. To bridge this gap, we conducted a comprehensive human study involving interviews with 26 guide dog handlers, four white cane users, nine guide dog trainers, and one O\&M trainer, along with 15+ hours of observing guide dog-assisted walking. After de-identification, we open-sourced the dataset to promote human-centered development and informed decision-making for assistive systems for BLV people. Building on insights from this formative study, we developed GuideNav, a vision-only, teach-and-repeat navigation system. Inspired by how guide dogs are trained and assist their handlers, GuideNav autonomously repeats a path demonstrated by a sighted person using a robot. Specifically, the system constructs a topological representation of the taught route, integrates visual place recognition with temporal filtering, and employs a relative pose estimator to compute navigation actions - all without relying on costly, heavy, power-hungry sensors such as LiDAR. In field tests, GuideNav consistently achieved kilometer-scale route following across five outdoor environments, maintaining reliability despite noticeable scene variations between teach and repeat runs. A user study with 3 guide dog handlers and 1 guide dog trainer further confirmed the system's feasibility, marking (to our knowledge) the first demonstration of a quadruped mobile system retrieving a path in a manner comparable to guide dogs.

Kuniaki Saito, Donghyun Kim, Kwanyong Park et al.

An image captioning model flexibly switching its language pattern, e.g., descriptiveness and length, should be useful since it can be applied to diverse applications. However, despite the dramatic improvement in generative vision-language models, fine-grained control over the properties of generated captions is not easy due to two reasons: (i) existing models are not given the properties as a condition during training and (ii) existing models cannot smoothly transition its language pattern from one state to the other. Given this challenge, we propose a new approach, CaptionSmiths, to acquire a single captioning model that can handle diverse language patterns. First, our approach quantifies three properties of each caption, length, descriptiveness, and uniqueness of a word, as continuous scalar values, without human annotation. Given the values, we represent the conditioning via interpolation between two endpoint vectors corresponding to the extreme states, e.g., one for a very short caption and one for a very long caption. Empirical results demonstrate that the resulting model can smoothly change the properties of the output captions and show higher lexical alignment than baselines. For instance, CaptionSmiths reduces the error in controlling caption length by 506\% despite better lexical alignment. Code will be available on https://github.com/omron-sinicx/captionsmiths.

Jongwoo Park, Kanchana Ranasinghe, Kumara Kahatapitiya et al.

Long-form videos that span across wide temporal intervals are highly information redundant and contain multiple distinct events or entities that are often loosely related. Therefore, when performing long-form video question answering (LVQA), all information necessary to generate a correct response can often be contained within a small subset of frames. Recent literature leverage large language models (LLMs) in LVQA benchmarks, achieving exceptional performance, while relying on vision language models (VLMs) to convert all visual content within videos into natural language. Such VLMs often independently caption a large number of frames uniformly sampled from long videos, which is not efficient and can mostly be redundant. Motivated by this inefficiency, we propose LVNet, a modular and training-free framework featuring a novel Hierarchical Keyframe Selector (HKS) that efficiently selects a minimal set of informative frames tailored to each question. LVNet's modularity allows easy integration with existing approaches for more efficient LVQA. We achieve state-of-the-art performance among similarly configured models across four benchmark LVQA datasets: EgoSchema, NExT-QA, IntentQA, VideoMME. The code can be found at https://github.com/jongwoopark7978/LVNet

Kuniaki Saito, Donghyun Kim, Piotr Teterwak et al.

Unsupervised domain adaptation (UDA) methods can dramatically improve generalization on unlabeled target domains. However, optimal hyper-parameter selection is critical to achieving high accuracy and avoiding negative transfer. Supervised hyper-parameter validation is not possible without labeled target data, which raises the question: How can we validate unsupervised adaptation techniques in a realistic way? We first empirically analyze existing criteria and demonstrate that they are not very effective for tuning hyper-parameters. Intuitively, a well-trained source classifier should embed target samples of the same class nearby, forming dense neighborhoods in feature space. Based on this assumption, we propose a novel unsupervised validation criterion that measures the density of soft neighborhoods by computing the entropy of the similarity distribution between points. Our criterion is simpler than competing validation methods, yet more effective; it can tune hyper-parameters and the number of training iterations in both image classification and semantic segmentation models. The code used for the paper will be available at \url{https://github.com/VisionLearningGroup/SND}.

Teakgyu Hong, Donghyun Kim, Mingi Ji et al.

Key information extraction (KIE) from document images requires understanding the contextual and spatial semantics of texts in two-dimensional (2D) space. Many recent studies try to solve the task by developing pre-trained language models focusing on combining visual features from document images with texts and their layout. On the other hand, this paper tackles the problem by going back to the basic: effective combination of text and layout. Specifically, we propose a pre-trained language model, named BROS (BERT Relying On Spatiality), that encodes relative positions of texts in 2D space and learns from unlabeled documents with area-masking strategy. With this optimized training scheme for understanding texts in 2D space, BROS shows comparable or better performance compared to previous methods on four KIE benchmarks (FUNSD, SROIE*, CORD, and SciTSR) without relying on visual features. This paper also reveals two real-world challenges in KIE tasks-(1) minimizing the error from incorrect text ordering and (2) efficient learning from fewer downstream examples-and demonstrates the superiority of BROS over previous methods. Code is available at https://github.com/clovaai/bros.

Kuniaki Saito, Donghyun Kim, Stan Sclaroff et al.

Unsupervised domain adaptation methods traditionally assume that all source categories are present in the target domain. In practice, little may be known about the category overlap between the two domains. While some methods address target settings with either partial or open-set categories, they assume that the particular setting is known a priori. We propose a more universally applicable domain adaptation framework that can handle arbitrary category shift, called Domain Adaptative Neighborhood Clustering via Entropy optimization (DANCE). DANCE combines two novel ideas: First, as we cannot fully rely on source categories to learn features discriminative for the target, we propose a novel neighborhood clustering technique to learn the structure of the target domain in a self-supervised way. Second, we use entropy-based feature alignment and rejection to align target features with the source, or reject them as unknown categories based on their entropy. We show through extensive experiments that DANCE outperforms baselines across open-set, open-partial and partial domain adaptation settings. Implementation is available at https://github.com/VisionLearningGroup/DANCE.

Donghyun Kim, Aws Khalil, Haewoon Nam et al.

Autonomous Vehicles (AV) and Advanced Driver Assistant Systems (ADAS) prioritize safety over comfort. The intertwining factors of safety and comfort emerge as pivotal elements in ensuring the effectiveness of Autonomous Driving (AD). Users often experience discomfort when AV or ADAS drive the vehicle on their behalf. Providing a personalized human-like AD experience, tailored to match users' unique driving styles while adhering to safety prerequisites, presents a significant opportunity to boost the acceptance of AVs. This paper proposes a novel approach, Neural Driving Style Transfer (NDST), inspired by Neural Style Transfer (NST), to address this issue. NDST integrates a Personalized Block (PB) into the conventional Baseline Driving Model (BDM), allowing for the transfer of a user's unique driving style while adhering to safety parameters. The PB serves as a self-configuring system, learning and adapting to an individual's driving behavior without requiring modifications to the BDM. This approach enables the personalization of AV models, aligning the driving style more closely with user preferences while ensuring baseline safety critical actuation. Two contrasting driving styles (Style A and Style B) were used to validate the proposed NDST methodology, demonstrating its efficacy in transferring personal driving styles to the AV system. Our work highlights the potential of NDST to enhance user comfort in AVs by providing a personalized and familiar driving experience. The findings affirm the feasibility of integrating NDST into existing AV frameworks to bridge the gap between safety and individualized driving styles, promoting wider acceptance and improved user experiences.

Jungmyung Wi, Hyunsoo Kim, Donghyun Kim

Text-to-image models produce images that align well with natural language prompts, but compositional generation has long been a central challenge. Models often struggle to satisfy multiple concepts within a single prompt, frequently omitting some concepts and resulting in partial success. Such failures highlight the difficulty of jointly optimizing multiple concepts during reward optimization, where competing concepts can interfere with one another. To address this limitation, we propose Correlation-Weighted Multi-Reward Optimization (\ours), a framework that leverages the correlation structure among concept rewards to adaptively weight each attribute concept in optimization. By accounting for interactions among concepts, \ours balances competing reward signals and emphasizes concepts that are partially satisfied yet inconsistently generated across samples, improving compositional generation. Specifically, we decompose multi-concept prompts into pre-defined concept groups (\eg, objects, attributes, and relations) and obtain reward signals from dedicated reward models for each concept. We then adaptively reweight these rewards, assigning higher weights to conflicting or hard-to-satisfy concepts using correlation-based difficulty estimation. By focusing optimization on the most challenging concepts within each group, \ours encourages the model to consistently satisfy all requested attributes simultaneously. We apply our approach to train state-of-the-art diffusion models, SD3.5 and FLUX.1-dev, and demonstrate consistent improvements on challenging multi-concept benchmarks, including ConceptMix, GenEval 2, and T2I-CompBench.

Yongwoo Kim, Sungmin Cha, Hyunsoo Kim et al.

With the increasing versatility of text-to-image diffusion models, the ability to selectively erase undesirable concepts (e.g., harmful content) has become indispensable. However, existing concept erasure approaches primarily focus on removing unsafe concepts without providing guidance toward corresponding safe alternatives, which often leads to failure in preserving the structural and semantic consistency between the original and erased generations. In this paper, we propose a novel framework, PAIRed Erasing (PAIR), which reframes concept erasure from simple removal to consistency-preserving semantic realignment using unsafe-safe pairs. We first generate safe counterparts from unsafe inputs while preserving structural and semantic fidelity, forming paired unsafe-safe multimodal data. Leveraging these pairs, we introduce two key components: (1) Paired Semantic Realignment, a guided objective that uses unsafe-safe pairs to explicitly map target concepts to semantically aligned safe anchors; and (2) Fisher-weighted Initialization for DoRA, which initializes parameter-efficient low-rank adaptation matrices using unsafe-safe pairs, encouraging the generation of safe alternatives while selectively suppressing unsafe concepts. Together, these components enable fine-grained erasure that removes only the targeted concepts while maintaining overall semantic consistency. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art baselines, achieving effective concept erasure while preserving structural integrity, semantic coherence, and generation quality.

Hochul Hwang, Sunjae Kwon, Yekyung Kim et al.

Safely navigating street intersections is a complex challenge for blind and low-vision individuals, as it requires a nuanced understanding of the surrounding context - a task heavily reliant on visual cues. Traditional methods for assisting in this decision-making process often fall short, lacking the ability to provide a comprehensive scene analysis and safety level. This paper introduces an innovative approach that leverages large multimodal models (LMMs) to interpret complex street crossing scenes, offering a potential advancement over conventional traffic signal recognition techniques. By generating a safety score and scene description in natural language, our method supports safe decision-making for the blind and low-vision individuals. We collected crosswalk intersection data that contains multiview egocentric images captured by a quadruped robot and annotated the images with corresponding safety scores based on our predefined safety score categorization. Grounded on the visual knowledge, extracted from images, and text prompt, we evaluate a large multimodal model for safety score prediction and scene description. Our findings highlight the reasoning and safety score prediction capabilities of a LMM, activated by various prompts, as a pathway to developing a trustworthy system, crucial for applications requiring reliable decision-making support.

Young Kyun Jang, Donghyun Kim, Zihang Meng et al.

Composed Image Retrieval (CIR) is a task that retrieves images similar to a query, based on a provided textual modification. Current techniques rely on supervised learning for CIR models using labeled triplets of the reference image, text, target image. These specific triplets are not as commonly available as simple image-text pairs, limiting the widespread use of CIR and its scalability. On the other hand, zero-shot CIR can be relatively easily trained with image-caption pairs without considering the image-to-image relation, but this approach tends to yield lower accuracy. We propose a new semi-supervised CIR approach where we search for a reference and its related target images in auxiliary data and learn our large language model-based Visual Delta Generator (VDG) to generate text describing the visual difference (i.e., visual delta) between the two. VDG, equipped with fluent language knowledge and being model agnostic, can generate pseudo triplets to boost the performance of CIR models. Our approach significantly improves the existing supervised learning approaches and achieves state-of-the-art results on the CIR benchmarks.