Sung-eui Yoon

Yoonki Cho, Woo Jae Kim, Seunghoon Hong et al.

Unsupervised person re-identification (re-ID) aims at learning discriminative representations for person retrieval from unlabeled data. Recent techniques accomplish this task by using pseudo-labels, but these labels are inherently noisy and deteriorate the accuracy. To overcome this problem, several pseudo-label refinement methods have been proposed, but they neglect the fine-grained local context essential for person re-ID. In this paper, we propose a novel Part-based Pseudo Label Refinement (PPLR) framework that reduces the label noise by employing the complementary relationship between global and part features. Specifically, we design a cross agreement score as the similarity of k-nearest neighbors between feature spaces to exploit the reliable complementary relationship. Based on the cross agreement, we refine pseudo-labels of global features by ensembling the predictions of part features, which collectively alleviate the noise in global feature clustering. We further refine pseudo-labels of part features by applying label smoothing according to the suitability of given labels for each part. Thanks to the reliable complementary information provided by the cross agreement score, our PPLR effectively reduces the influence of noisy labels and learns discriminative representations with rich local contexts. Extensive experimental results on Market-1501 and MSMT17 demonstrate the effectiveness of the proposed method over the state-of-the-art performance. The code is available at https://github.com/yoonkicho/PPLR.

Woo Jae Kim, Yoonki Cho, Junsik Jung et al.

Deep neural networks are susceptible to adversarial attacks due to the accumulation of perturbations in the feature level, and numerous works have boosted model robustness by deactivating the non-robust feature activations that cause model mispredictions. However, we claim that these malicious activations still contain discriminative cues and that with recalibration, they can capture additional useful information for correct model predictions. To this end, we propose a novel, easy-to-plugin approach named Feature Separation and Recalibration (FSR) that recalibrates the malicious, non-robust activations for more robust feature maps through Separation and Recalibration. The Separation part disentangles the input feature map into the robust feature with activations that help the model make correct predictions and the non-robust feature with activations that are responsible for model mispredictions upon adversarial attack. The Recalibration part then adjusts the non-robust activations to restore the potentially useful cues for model predictions. Extensive experiments verify the superiority of FSR compared to traditional deactivation techniques and demonstrate that it improves the robustness of existing adversarial training methods by up to 8.57% with small computational overhead. Codes are available at https://github.com/wkim97/FSR.

Woo Jae Kim, Seunghoon Hong, Sung-Eui Yoon

Adversarial attacks with improved transferability - the ability of an adversarial example crafted on a known model to also fool unknown models - have recently received much attention due to their practicality. Nevertheless, existing transferable attacks craft perturbations in a deterministic manner and often fail to fully explore the loss surface, thus falling into a poor local optimum and suffering from low transferability. To solve this problem, we propose Attentive-Diversity Attack (ADA), which disrupts diverse salient features in a stochastic manner to improve transferability. Primarily, we perturb the image attention to disrupt universal features shared by different models. Then, to effectively avoid poor local optima, we disrupt these features in a stochastic manner and explore the search space of transferable perturbations more exhaustively. More specifically, we use a generator to produce adversarial perturbations that each disturbs features in different ways depending on an input latent code. Extensive experimental evaluations demonstrate the effectiveness of our method, outperforming the transferability of state-of-the-art methods. Codes are available at https://github.com/wkim97/ADA.

Jumin Lee, Woobin Im, Sebin Lee et al.

In this paper, we learn a diffusion model to generate 3D data on a scene-scale. Specifically, our model crafts a 3D scene consisting of multiple objects, while recent diffusion research has focused on a single object. To realize our goal, we represent a scene with discrete class labels, i.e., categorical distribution, to assign multiple objects into semantic categories. Thus, we extend discrete diffusion models to learn scene-scale categorical distributions. In addition, we validate that a latent diffusion model can reduce computation costs for training and deploying. To the best of our knowledge, our work is the first to apply discrete and latent diffusion for 3D categorical data on a scene-scale. We further propose to perform semantic scene completion (SSC) by learning a conditional distribution using our diffusion model, where the condition is a partial observation in a sparse point cloud. In experiments, we empirically show that our diffusion models not only generate reasonable scenes, but also perform the scene completion task better than a discriminative model. Our code and models are available at https://github.com/zoomin-lee/scene-scale-diffusion

Guoyuan An, Woo Jae Kim, Saelyne Yang et al.

This paper introduces the first two pixel retrieval benchmarks. Pixel retrieval is segmented instance retrieval. Like semantic segmentation extends classification to the pixel level, pixel retrieval is an extension of image retrieval and offers information about which pixels are related to the query object. In addition to retrieving images for the given query, it helps users quickly identify the query object in true positive images and exclude false positive images by denoting the correlated pixels. Our user study results show pixel-level annotation can significantly improve the user experience. Compared with semantic and instance segmentation, pixel retrieval requires a fine-grained recognition capability for variable-granularity targets. To this end, we propose pixel retrieval benchmarks named PROxford and PRParis, which are based on the widely used image retrieval datasets, ROxford and RParis. Three professional annotators label 5,942 images with two rounds of double-checking and refinement. Furthermore, we conduct extensive experiments and analysis on the SOTA methods in image search, image matching, detection, segmentation, and dense matching using our pixel retrieval benchmarks. Results show that the pixel retrieval task is challenging to these approaches and distinctive from existing problems, suggesting that further research can advance the content-based pixel-retrieval and thus user search experience. The datasets can be downloaded from \href{https://github.com/anguoyuan/Pixel_retrieval-Segmented_instance_retrieval}{this link}.

Woobin Im, Sebin Lee, Sung-Eui Yoon

A training pipeline for optical flow CNNs consists of a pretraining stage on a synthetic dataset followed by a fine tuning stage on a target dataset. However, obtaining ground truth flows from a target video requires a tremendous effort. This paper proposes a practical fine tuning method to adapt a pretrained model to a target dataset without ground truth flows, which has not been explored extensively. Specifically, we propose a flow supervisor for self-supervision, which consists of parameter separation and a student output connection. This design is aimed at stable convergence and better accuracy over conventional self-supervision methods which are unstable on the fine tuning task. Experimental results show the effectiveness of our method compared to different self-supervision methods for semi-supervised learning. In addition, we achieve meaningful improvements over state-of-the-art optical flow models on Sintel and KITTI benchmarks by exploiting additional unlabeled datasets. Code is available at https://github.com/iwbn/flow-supervisor.

Pei Wang, Danna Xue, Yu Zhu et al.

Recovering clear structures from severely blurry inputs is a challenging problem due to the large movements between the camera and the scene. Although some works apply segmentation maps on human face images for deblurring, they cannot handle natural scenes because objects and degradation are more complex, and inaccurate segmentation maps lead to a loss of details. For general scene deblurring, the feature space of the blurry image and corresponding sharp image under the high-level vision task is closer, which inspires us to rely on other tasks (e.g. classification) to learn a comprehensive prior in severe blur removal cases. We propose a cross-level feature learning strategy based on knowledge distillation to learn the priors, which include global contexts and sharp local structures for recovering potential details. In addition, we propose a semantic prior embedding layer with multi-level aggregation and semantic attention transformation to integrate the priors effectively. We introduce the proposed priors to various models, including the UNet and other mainstream deblurring baselines, leading to better performance on severe blur removal. Extensive experiments on natural image deblurring benchmarks and real-world images, such as GoPro and RealBlur datasets, demonstrate our method's effectiveness and generalization ability.

Xu Yin, Dongbo Min, Yuchi Huo et al.

With increasing demands for high-quality semantic segmentation in the industry, hard-distinguishing semantic boundaries have posed a significant threat to existing solutions. Inspired by real-life experience, i.e., combining varied observations contributes to higher visual recognition confidence, we present the equipotential learning (EPL) method. This novel module transfers the predicted/ground-truth semantic labels to a self-defined potential domain to learn and infer decision boundaries along customized directions. The conversion to the potential domain is implemented via a lightweight differentiable anisotropic convolution without incurring any parameter overhead. Besides, the designed two loss functions, the point loss and the equipotential line loss implement anisotropic field regression and category-level contour learning, respectively, enhancing prediction consistencies in the inter/intra-class boundary areas. More importantly, EPL is agnostic to network architectures, and thus it can be plugged into most existing segmentation models. This paper is the first attempt to address the boundary segmentation problem with field regression and contour learning. Meaningful performance improvements on Pascal Voc 2012 and Cityscapes demonstrate that the proposed EPL module can benefit the off-the-shelf fully convolutional network models when recognizing semantic boundary areas. Besides, intensive comparisons and analysis show the favorable merits of EPL for distinguishing semantically-similar and irregular-shaped categories.

Sebin Lee, Jumin Lee, Taeyeon Kim et al.

Rapidly-exploring random trees (RRTs) have been widely adopted for robot motion planning due to their robustness and theoretical guarantees. However, existing RRT-based planners require explicit goal configurations specified as numerical joint angles, while many practical applications provide goal specifications through visual observations such as images or demonstration videos where precise goal configurations are unavailable. In this paper, we propose visual-RRT (vRRT), a motion planner that enables visual-goal planning by unifying gradient-based exploitation from differentiable robot rendering with sampling-based exploration from RRTs. We further introduce (i) a frontier-based exploration-exploitation strategy that adaptively prioritizes visually promising search regions, and (ii) inertial gradient tree expansion that inherits optimization states across tree branches for momentum-consistent gradient exploitation. Extensive experiments across various robot manipulators including Franka, UR5e, and Fetch demonstrate that vRRT achieves effective visual-goal planning in both simulated and real-world settings, bridging the gap between sampling-based planning and vision-centric robot applications. Our code is available at https://sgvr.kaist.ac.kr/Visual-RRT.

Kyuyeon Kim, Junsik Jung, Woo Jae Kim et al.

Humans can easily imagine a scene from auditory information based on their prior knowledge of audio-visual events. In this paper, we mimic this innate human ability in deep learning models to improve the quality of video inpainting. To implement the prior knowledge, we first train the audio-visual network, which learns the correspondence between auditory and visual information. Then, the audio-visual network is employed as a guider that conveys the prior knowledge of audio-visual correspondence to the video inpainting network. This prior knowledge is transferred through our proposed two novel losses: audio-visual attention loss and audio-visual pseudo-class consistency loss. These two losses further improve the performance of the video inpainting by encouraging the inpainting result to have a high correspondence to its synchronized audio. Experimental results demonstrate that our proposed method can restore a wider domain of video scenes and is particularly effective when the sounding object in the scene is partially blinded.

Jaeyoon Kim, Yoonki Cho, Sung-Eui Yoon

Visual Place Recognition (VPR) has advanced significantly with high-capacity foundation models like DINOv2, achieving remarkable performance. Nonetheless, their substantial computational cost makes deployment on resource-constrained devices impractical. In this paper, we introduce an efficient asymmetric VPR framework that incorporates a high-capacity gallery model for offline feature extraction with a lightweight query network for online processing. A key challenge in this setting is ensuring compatibility between these heterogeneous networks, which conventional approaches address through computationally expensive k-NN-based compatible training. To overcome this, we propose a geographical memory bank that structures gallery features using geolocation metadata inherent in VPR databases, eliminating the need for exhaustive k-NN computations. Additionally, we introduce an implicit embedding augmentation technique that enhances the query network to model feature variations despite its limited capacity. Extensive experiments demonstrate that our method not only significantly reduces computational costs but also outperforms existing asymmetric retrieval techniques, establishing a new aspect for VPR in resource-limited environments. The code is available at https://github.com/jaeyoon1603/AsymVPR

Guoyuan An, Juhyung Seon, Inkyu An et al.

This paper presents an innovative approach to enhancing explainable image retrieval, particularly in situations where a fine-tuning set is unavailable. The widely-used SPatial verification (SP) method, despite its efficacy, relies on a spatial model and the hypothesis-testing strategy for instance recognition, leading to inherent limitations, including the assumption of planar structures and neglect of topological relations among features. To address these shortcomings, we introduce a pioneering technique that replaces the spatial model with a topological one within the RANSAC process. We propose bio-inspired saccade and fovea functions to verify the topological consistency among features, effectively circumventing the issues associated with SP's spatial model. Our experimental results demonstrate that our method significantly outperforms SP, achieving state-of-the-art performance in non-fine-tuning retrieval. Furthermore, our approach can enhance performance when used in conjunction with fine-tuned features. Importantly, our method retains high explainability and is lightweight, offering a practical and adaptable solution for a variety of real-world applications.

Xu Yin, Fei Pan, Guoyuan An et al.

Existing open-set recognition (OSR) studies typically assume that each image contains only one class label, with the unknown test set (negative) having a disjoint label space from the known test set (positive), a scenario referred to as full-label shift. This paper introduces the mixed OSR problem, where test images contain multiple class semantics, with both known and unknown classes co-occurring in the negatives, leading to a more complex super-label shift that better reflects real-world scenarios. To tackle this challenge, we propose the OpenSlot framework, based on object-centric learning, which uses slot features to represent diverse class semantics and generate class predictions. The proposed anti-noise slot (ANS) technique helps mitigate the impact of noise (invalid or background) slots during classification training, addressing the semantic misalignment between class predictions and ground truth. We evaluate OpenSlot on both mixed and conventional OSR benchmarks. Without elaborate designs, our method not only excels existing approaches in detecting super-label shifts across OSR tasks, but also achieves state-of-the-art performance on conventional benchmarks. Meanwhile, OpenSlot can localize class objects without using bounding boxes during training, demonstrating competitive performance in open-set object detection and potential for generalization.

Jumin Lee, Siyeong Lee, Namil Kim et al.

Long-tail distributions in driving datasets pose a fundamental challenge for 3D perception, as rare classes exhibit substantial intra-class diversity yet available samples cover this variation space only sparsely. Existing instance augmentation methods based on copy-paste or asset libraries improve rare-class exposure but are often limited in fine-grained diversity and scene-context placement. We propose VERIA, an image-first multimodal augmentation framework that synthesizes synchronized RGB--LiDAR instances using off-the-shelf foundation models and curates them with sequential semantic and geometric verification. This verification-centric design tends to select instances that better match real LiDAR statistics while spanning a wider range of intra-class variation. Stage-wise yield decomposition provides a log-based diagnostic of pipeline reliability. On nuScenes and Lyft, VERIA improves rare-class 3D object detection in both LiDAR-only and multimodal settings. Our code is available at https://sgvr.kaist.ac.kr/VERIA/.

Jumin Lee, Sebin Lee, Changho Jo et al.

We present "SemCity," a 3D diffusion model for semantic scene generation in real-world outdoor environments. Most 3D diffusion models focus on generating a single object, synthetic indoor scenes, or synthetic outdoor scenes, while the generation of real-world outdoor scenes is rarely addressed. In this paper, we concentrate on generating a real-outdoor scene through learning a diffusion model on a real-world outdoor dataset. In contrast to synthetic data, real-outdoor datasets often contain more empty spaces due to sensor limitations, causing challenges in learning real-outdoor distributions. To address this issue, we exploit a triplane representation as a proxy form of scene distributions to be learned by our diffusion model. Furthermore, we propose a triplane manipulation that integrates seamlessly with our triplane diffusion model. The manipulation improves our diffusion model's applicability in a variety of downstream tasks related to outdoor scene generation such as scene inpainting, scene outpainting, and semantic scene completion refinements. In experimental results, we demonstrate that our triplane diffusion model shows meaningful generation results compared with existing work in a real-outdoor dataset, SemanticKITTI. We also show our triplane manipulation facilitates seamlessly adding, removing, or modifying objects within a scene. Further, it also enables the expansion of scenes toward a city-level scale. Finally, we evaluate our method on semantic scene completion refinements where our diffusion model enhances predictions of semantic scene completion networks by learning scene distribution. Our code is available at https://github.com/zoomin-lee/SemCity.

Woobin Im, Geonho Cha, Sebin Lee et al.

This paper presents a novel approach for reconstructing dynamic radiance fields from monocular videos. We integrate kinematics with dynamic radiance fields, bridging the gap between the sparse nature of monocular videos and the real-world physics. Our method introduces the kinematic field, capturing motion through kinematic quantities: velocity, acceleration, and jerk. The kinematic field is jointly learned with the dynamic radiance field by minimizing the photometric loss without motion ground truth. We further augment our method with physics-driven regularizers grounded in kinematics. We propose physics-driven regularizers that ensure the physical validity of predicted kinematic quantities, including advective acceleration and jerk. Additionally, we control the motion trajectory based on rigidity equations formed with the predicted kinematic quantities. In experiments, our method outperforms the state-of-the-arts by capturing physical motion patterns within challenging real-world monocular videos.

Youngju Na, Woo Jae Kim, Kyu Beom Han et al.

Generalizable neural implicit surface reconstruction aims to obtain an accurate underlying geometry given a limited number of multi-view images from unseen scenes. However, existing methods select only informative and relevant views using predefined scores for training and testing phases. This constraint renders the model impractical in real-world scenarios, where the availability of favorable combinations cannot always be ensured. We introduce and validate a view-combination score to indicate the effectiveness of the input view combination. We observe that previous methods output degenerate solutions under arbitrary and unfavorable sets. Building upon this finding, we propose UFORecon, a robust view-combination generalizable surface reconstruction framework. To achieve this, we apply cross-view matching transformers to model interactions between source images and build correlation frustums to capture global correlations. Additionally, we explicitly encode pairwise feature similarities as view-consistent priors. Our proposed framework significantly outperforms previous methods in terms of view-combination generalizability and also in the conventional generalizable protocol trained with favorable view-combinations. The code is available at https://github.com/Youngju-Na/UFORecon.

Kyu Beom Han, Olivia G. Odenthal, Woo Jae Kim et al.

Auxiliary features such as geometric buffers (G-buffers) and path descriptors (P-buffers) have been shown to significantly improve Monte Carlo (MC) denoising. However, recent approaches implicitly learn to exploit auxiliary features for denoising, which could lead to insufficient utilization of each type of auxiliary features. To overcome such an issue, we propose a denoising framework that relies on an explicit pixel-wise guidance for utilizing auxiliary features. First, we train two denoisers, each trained by a different auxiliary feature (i.e., G-buffers or P-buffers). Then we design our ensembling network to obtain per-pixel ensembling weight maps, which represent pixel-wise guidance for which auxiliary feature should be dominant at reconstructing each individual pixel and use them to ensemble the two denoised results of our denosiers. We also propagate our pixel-wise guidance to the denoisers by jointly training the denoisers and the ensembling network, further guiding the denoisers to focus on regions where G-buffers or P-buffers are relatively important for denoising. Our result and show considerable improvement in denoising performance compared to the baseline denoising model using both G-buffers and P-buffers.

Xiangchen Liu, Hanghan Zheng, Jeil Jeong et al.

Vision-language Navigation (VLN) requires an agent to understand visual observations and language instructions to navigate in unseen environments. Most existing approaches rely on static scene assumptions and struggle to generalize in dynamic, real-world scenarios. To address this challenge, we propose DyGeoVLN, a dynamic geometry-aware VLN framework. Our method infuses a dynamic geometry foundation model into the VLN framework through cross-branch feature fusion to enable explicit 3D spatial representation and visual-semantic reasoning. To efficiently compress historical token information in long-horizon, dynamic navigation, we further introduce a novel pose-free and adaptive-resolution token-pruning strategy. This strategy can remove spatio-temporal redundant tokens to reduce inference cost. Extensive experiments demonstrate that our approach achieves state-of-the-art performance on multiple benchmarks and exhibits strong robustness in real-world environments.

Yoonki Cho, Jaeyoon Kim, Woo Jae Kim et al.

Domain generalizable person re-identification (DG re-ID) aims to learn discriminative representations that are robust to distributional shifts. While data augmentation is a straightforward solution to improve generalization, certain augmentations exhibit a polarized effect in this task, enhancing in-distribution performance while deteriorating out-of-distribution performance. In this paper, we investigate this phenomenon and reveal that it leads to sparse representation spaces with reduced uniformity. To address this issue, we propose a novel framework, Balancing Alignment and Uniformity (BAU), which effectively mitigates this effect by maintaining a balance between alignment and uniformity. Specifically, BAU incorporates alignment and uniformity losses applied to both original and augmented images and integrates a weighting strategy to assess the reliability of augmented samples, further improving the alignment loss. Additionally, we introduce a domain-specific uniformity loss that promotes uniformity within each source domain, thereby enhancing the learning of domain-invariant features. Extensive experimental results demonstrate that BAU effectively exploits the advantages of data augmentation, which previous studies could not fully utilize, and achieves state-of-the-art performance without requiring complex training procedures. The code is available at \url{https://github.com/yoonkicho/BAU}.

Jinhwan Seo, Yoonki Cho, Junhyug Noh et al.

In this technical report, we introduce a framework to address Grounded Video Question Answering (GVQA) task for the ICCV 2025 Perception Test Challenge. The GVQA task demands robust multimodal models capable of complex reasoning over video content, grounding the resulting answers visually, and tracking the referenced objects temporally. To achieve this capability, our proposed approach decomposes the GVQA task into a three-stage pipeline: (1) Video Reasoning \& QA, (2) Spatio-temporal Grounding and (3) Tracking. Our key contribution is the introduction of a trigger moment, derived from our proposed CORTEX prompt, which pinpoints the single most visible frame of a target object to serve as a robust anchor for grounding and tracking. To this end, we achieve the HOTA score of 0.4968, which marks a significant improvement over the previous year's winning score of 0.2704 on GVQA task.

Youngju Na, Jaeseong Yun, Soohyun Ryu et al.

While 3D Gaussian splatting has emerged as a powerful paradigm, it fundamentally fails to model transparency such as glass panels. The core challenge lies in decoupling the intertwined radiance contributions from transparent interfaces and the transmitted geometry observed through the glass. We present GLINT, a framework that models scene-scale transparency through explicit decomposed Gaussian representation. GLINT reconstructs the primary interface and models reflected and transmitted radiance separately, enabling consistent radiance transport. During optimization, GLINT bootstraps transparency localization from geometry-separation cues induced by the decomposition, together with geometry and material priors from a pre-trained video relighting model. Extensive experiments demonstrate consistent improvements over prior methods for reconstructing complex transparent scenes.

Andrew Jeong, Jaemin Kim, Sebin Lee et al.

Robotic manipulation involves kinematic and semantic transitions that are inherently coupled via underlying actions. However, existing approaches plan within either semantic or latent space without explicitly aligning these cross-modal transitions. To address this, we propose CLaD, a framework that models how proprioceptive and semantic states jointly evolve under actions through asymmetric cross-attention that allows kinematic transitions to query semantic ones. CLaD predicts grounded latent foresights via self-supervised objectives with EMA target encoders and auxiliary reconstruction losses, preventing representation collapse while anchoring predictions to observable states. Predicted foresights are modulated with observations to condition a diffusion policy for action generation. On LIBERO-LONG benchmark, CLaD achieves 94.7\% success rate, competitive with large VLAs with significantly fewer parameters.

Joonsung Jeon, Woo Jae Kim, Suhyeon Ha et al.

Latent diffusion models have achieved remarkable success in high-fidelity text-to-image generation, but their tendency to memorize training data raises critical privacy and intellectual property concerns. Membership inference attacks (MIAs) provide a principled way to audit such memorization by determining whether a given sample was included in training. However, existing approaches assume access to ground-truth captions. This assumption fails in realistic scenarios where only images are available and their textual annotations remain undisclosed, rendering prior methods ineffective when substituted with vision-language model (VLM) captions. In this work, we propose MoFit, a caption-free MIA framework that constructs synthetic conditioning inputs that are explicitly overfitted to the target model's generative manifold. Given a query image, MoFit proceeds in two stages: (i) model-fitted surrogate optimization, where a perturbation applied to the image is optimized to construct a surrogate in regions of the model's unconditional prior learned from member samples, and (ii) surrogate-driven embedding extraction, where a model-fitted embedding is derived from the surrogate and then used as a mismatched condition for the query image. This embedding amplifies conditional loss responses for member samples while leaving hold-outs relatively less affected, thereby enhancing separability in the absence of ground-truth captions. Our comprehensive experiments across multiple datasets and diffusion models demonstrate that MoFit consistently outperforms prior VLM-conditioned baselines and achieves performance competitive with caption-dependent methods.

Minsung Yoon, Jeil Jeong, Sung-Eui Yoon

Skateboards offer a compact and efficient means of transportation as a type of personal mobility device. However, controlling them with legged robots poses several challenges for policy learning due to perception-driven interactions and multi-modal control objectives across distinct skateboarding phases. To address these challenges, we introduce Phase-Aware Policy Learning (PAPL), a reinforcement-learning framework tailored for skateboarding with quadruped robots. PAPL leverages the cyclic nature of skateboarding by integrating phase-conditioned Feature-wise Linear Modulation layers into actor and critic networks, enabling a unified policy that captures phase-dependent behaviors while sharing robot-specific knowledge across phases. Our evaluations in simulation validate command-tracking accuracy and conduct ablation studies quantifying each component's contribution. We also compare locomotion efficiency against leg and wheel-leg baselines and show real-world transferability.

Youngju Na, Taeyeon Kim, Jumin Lee et al.

While generalizable 3D Gaussian splatting enables efficient, high-quality rendering of unseen scenes, it heavily depends on precise camera poses for accurate geometry. In real-world scenarios, obtaining accurate poses is challenging, leading to noisy pose estimates and geometric misalignments. To address this, we introduce SHARE, a pose-free, feed-forward Gaussian splatting framework that overcomes these ambiguities by joint shape and camera rays estimation. Instead of relying on explicit 3D transformations, SHARE builds a pose-aware canonical volume representation that seamlessly integrates multi-view information, reducing misalignment caused by inaccurate pose estimates. Additionally, anchor-aligned Gaussian prediction enhances scene reconstruction by refining local geometry around coarse anchors, allowing for more precise Gaussian placement. Extensive experiments on diverse real-world datasets show that our method achieves robust performance in pose-free generalizable Gaussian splatting. Code is avilable at https://github.com/youngju-na/SHARE

Taeyeon Kim, Youngju Na, Jumin Lee et al.

Transferring articulated motion from monocular videos to rigged 3D characters is challenging due to pose ambiguity in 2D observations and morphological differences between source and target. Existing approaches often follow a reconstruct-then-retarget paradigm, tying transfer quality to intermediate 3D reconstruction and limiting applicability to categories with parametric templates. We propose MorphGS, a framework that formulates motion retargeting as a target-driven analysis-by-synthesis problem, directly optimizing target morphology and pose through image-space supervision. A rig-coupled morphology parameterization factorizes character identity from time-varying joint rotations, while dense 2D-3D correspondences and synthesized views provide complementary structural and multi-view guidance. Experiments on synthetic benchmarks and in-the-wild videos show consistent improvements over baselines.

Woo Jae Kim, Kyu Beom Han, Yoonki Cho et al.

As Neural Radiance Fields (NeRFs) have emerged as a powerful tool for 3D scene representation and novel view synthesis, protecting their intellectual property (IP) from unauthorized use is becoming increasingly crucial. In this work, we aim to protect the IP of NeRFs by injecting adversarial perturbations that disrupt their unauthorized applications. However, perturbing the 3D geometry of NeRFs can easily deform the underlying scene structure and thus substantially degrade the rendering quality, which has led existing attempts to avoid geometric perturbations or restrict them to explicit spaces like meshes. To overcome this limitation, we introduce a learnable sensitivity to quantify the spatially varying impact of geometric perturbations on rendering quality. Building upon this, we propose AegisRF, a novel framework that consists of a Perturbation Field, which injects adversarial perturbations into the pre-rendering outputs (color and volume density) of NeRF models to fool an unauthorized downstream target model, and a Sensitivity Field, which learns the sensitivity to adaptively constrain geometric perturbations, preserving rendering quality while disrupting unauthorized use. Our experimental evaluations demonstrate the generalized applicability of AegisRF across diverse downstream tasks and modalities, including multi-view image classification and voxel-based 3D localization, while maintaining high visual fidelity. Codes are available at https://github.com/wkim97/AegisRF.

Kyu Beom Han, Jaeyoon Kim, Woo Jae Kim et al.

Inverse rendering with Gaussian Splatting has advanced rapidly, but accurately disentangling material properties from complex global illumination effects, particularly indirect illumination, remains a major challenge. Existing methods often query indirect radiance from Gaussian primitives pre-trained for novel-view synthesis. However, these pre-trained Gaussian primitives are supervised only towards limited training viewpoints, thus lack supervision for modeling indirect radiances from unobserved views. To address this issue, we introduce radiometric consistency, a novel physically-based constraint that provides supervision towards unobserved views by minimizing the residual between each Gaussian primitive's learned radiance and its physically-based rendered counterpart. Minimizing the residual for unobserved views establishes a self-correcting feedback loop that provides supervision from both physically-based rendering and novel-view synthesis, enabling accurate modeling of inter-reflection. We then propose Radiometrically Consistent Gaussian Surfels (RadioGS), an inverse rendering framework built upon our principle by efficiently integrating radiometric consistency by utilizing Gaussian surfels and 2D Gaussian ray tracing. We further propose a finetuning-based relighting strategy that adapts Gaussian surfel radiances to new illuminations within minutes, achieving low rendering cost (<10ms). Extensive experiments on existing inverse rendering benchmarks show that RadioGS outperforms existing Gaussian-based methods in inverse rendering, while retaining the computational efficiency.

Jeil Jeong, Minsung Yoon, Seokryun Choi et al.

Navigating quadruped robots in unstructured 3D environments poses significant challenges, requiring goal-directed motion, effective exploration to escape from local minima, and posture adaptation to traverse narrow, height-constrained spaces. Conventional approaches employ a sequential mapping-planning pipeline but suffer from accumulated perception errors and high computational overhead, restricting their applicability on resource-constrained platforms. To address these challenges, we propose Hierarchical Posture-Adaptive Navigation (HiPAN), a framework that operates directly on onboard depth images at deployment. HiPAN adopts a hierarchical design: a high-level policy generates strategic navigation commands (planar velocity and body posture), which are executed by a low-level, posture-adaptive locomotion controller. To mitigate myopic behaviors and facilitate long-horizon navigation, we introduce Path-Guided Curriculum Learning, which progressively extends the navigation horizon from reactive obstacle avoidance to strategic navigation. In simulation, HiPAN achieves higher navigation success rates and greater path efficiency than classical reactive planners and end-to-end baselines, while real-world experiments further validate its applicability across diverse, unstructured 3D environments.

Joonsung Jeon, Woo Jae Kim, Suhyeon Ha et al.

The outstanding capability of diffusion models in generating high-quality images poses significant threats when misused by adversaries. In particular, we assume malicious adversaries exploiting diffusion models for inpainting tasks, such as replacing a specific region with a celebrity. While existing methods for protecting images from manipulation in diffusion-based generative models have primarily focused on image-to-image and text-to-image tasks, the challenge of preventing unauthorized inpainting has been rarely addressed, often resulting in suboptimal protection performance. To mitigate inpainting abuses, we propose ADVPAINT, a novel defensive framework that generates adversarial perturbations that effectively disrupt the adversary's inpainting tasks. ADVPAINT targets the self- and cross-attention blocks in a target diffusion inpainting model to distract semantic understanding and prompt interactions during image generation. ADVPAINT also employs a two-stage perturbation strategy, dividing the perturbation region based on an enlarged bounding box around the object, enhancing robustness across diverse masks of varying shapes and sizes. Our experimental results demonstrate that ADVPAINT's perturbations are highly effective in disrupting the adversary's inpainting tasks, outperforming existing methods; ADVPAINT attains over a 100-point increase in FID and substantial decreases in precision.

Chanmi Lee, Minsung Yoon, Woojae Kim et al.

Neural network-based visuomotor policies enable robots to perform manipulation tasks but remain susceptible to perceptual attacks. For example, conventional 2D adversarial patches are effective under fixed-camera setups, where appearance is relatively consistent; however, their efficacy often diminishes under dynamic viewpoints from moving cameras, such as wrist-mounted setups, due to perspective distortions. To proactively investigate potential vulnerabilities beyond 2D patches, this work proposes a viewpoint-consistent adversarial texture optimization method for 3D objects through differentiable rendering. As optimization strategies, we employ Expectation over Transformation (EOT) with a Coarse-to-Fine (C2F) curriculum, exploiting distance-dependent frequency characteristics to induce textures effective across varying camera-object distances. We further integrate saliency-guided perturbations to redirect policy attention and design a targeted loss that persistently drives robots toward adversarial objects. Our comprehensive experiments show that the proposed method is effective under various environmental conditions, while confirming its black-box transferability and real-world applicability.

Xu Yin, Min-Sung Yoon, Yuchi Huo et al.

Task execution for object rearrangement could be challenged by Task-Level Perturbations (TLP), i.e., unexpected object additions, removals, and displacements that can disrupt underlying visual policies and fundamentally compromise task feasibility and progress. To address these challenges, we present LangPert, a language-based framework designed to detect and mitigate TLP situations in tabletop rearrangement tasks. LangPert integrates a Visual Language Model (VLM) to comprehensively monitor policy's skill execution and environmental TLP, while leveraging the Hierarchical Chain-of-Thought (HCoT) reasoning mechanism to enhance the Large Language Model (LLM)'s contextual understanding and generate adaptive, corrective skill-execution plans. Our experimental results demonstrate that LangPert handles diverse TLP situations more effectively than baseline methods, achieving higher task completion rates, improved execution efficiency, and potential generalization to unseen scenarios.

Jaeyoon Kim, Yoonki Cho, Taeyoung Kim et al.

Nearest neighbor (NN) graph based visual re-ranking has emerged as a powerful approach for improving retrieval accuracy, offering the advantages of effectively exploring high-dimensional manifolds without requiring additional fine-tuning. However, the effectiveness of NN graph-based re-ranking is fundamentally constrained by the quality of its edge connectivity, as incorrect connections between dissimilar (negative) images frequently occur. This is known as a noisy edge problem, which hinders the re-ranking performance of existing techniques and limits their potential. To remedy this issue, we propose a complementary denoising method based on Continuous Conditional Random Fields (C-CRF) that leverages statistical distances derived from similarity-based distributions. As a pre-processing step for enhancing NN graph-based retrieval, our approach constructs fully connected cliques around each anchor image and employs a novel statistical distance metric to robustly alleviate noisy edges before re-ranking while achieving efficient processing through offline computation. Extensive experimental results demonstrate that our method consistently improves three different NN graph-based re-ranking approaches, yielding significant gains in retrieval accuracy.

Xu Yin, Woobin Im, Dongbo Min et al.

Generating reliable pseudo masks from image-level labels is challenging in the weakly supervised semantic segmentation (WSSS) task due to the lack of spatial information. Prevalent class activation map (CAM)-based solutions are challenged to discriminate the foreground (FG) objects from the suspicious background (BG) pixels (a.k.a. co-occurring) and learn the integral object regions. This paper proposes a simple fine-grained background representation (FBR) method to discover and represent diverse BG semantics and address the co-occurring problems. We abandon using the class prototype or pixel-level features for BG representation. Instead, we develop a novel primitive, negative region of interest (NROI), to capture the fine-grained BG semantic information and conduct the pixel-to-NROI contrast to distinguish the confusing BG pixels. We also present an active sampling strategy to mine the FG negatives on-the-fly, enabling efficient pixel-to-pixel intra-foreground contrastive learning to activate the entire object region. Thanks to the simplicity of design and convenience in use, our proposed method can be seamlessly plugged into various models, yielding new state-of-the-art results under various WSSS settings across benchmarks. Leveraging solely image-level (I) labels as supervision, our method achieves 73.2 mIoU and 45.6 mIoU segmentation results on Pascal Voc and MS COCO test sets, respectively. Furthermore, by incorporating saliency maps as an additional supervision signal (I+S), we attain 74.9 mIoU on Pascal Voc test set. Concurrently, our FBR approach demonstrates meaningful performance gains in weakly-supervised instance segmentation (WSIS) tasks, showcasing its robustness and strong generalization capabilities across diverse domains.

Guoyuan An, Yuchi Huo, Sung-Eui Yoon

This paper presents a novel method designed to enhance the efficiency and accuracy of both image retrieval and pixel retrieval. Traditional diffusion methods struggle to propagate spatial information effectively in conventional graphs due to their reliance on scalar edge weights. To overcome this limitation, we introduce a hypergraph-based framework, uniquely capable of efficiently propagating spatial information using local features during query time, thereby accurately retrieving and localizing objects within a database. Additionally, we innovatively utilize the structural information of the image graph through a technique we term "community selection". This approach allows for the assessment of the initial search result's uncertainty and facilitates an optimal balance between accuracy and speed. This is particularly crucial in real-world applications where such trade-offs are often necessary. Our experimental results, conducted on the (P)ROxford and (P)RParis datasets, demonstrate the significant superiority of our method over existing diffusion techniques. We achieve state-of-the-art (SOTA) accuracy in both image-level and pixel-level retrieval, while also maintaining impressive processing speed. This dual achievement underscores the effectiveness of our hypergraph-based framework and community selection technique, marking a notable advancement in the field of content-based image retrieval.

Juhyeong Seon, Woobin Im, Sebin Lee et al.

Audio-visual segmentation (AVS) aims to segment sound sources in the video sequence, requiring a pixel-level understanding of audio-visual correspondence. As the Segment Anything Model (SAM) has strongly impacted extensive fields of dense prediction problems, prior works have investigated the introduction of SAM into AVS with audio as a new modality of the prompt. Nevertheless, constrained by SAM's single-frame segmentation scheme, the temporal context across multiple frames of audio-visual data remains insufficiently utilized. To this end, we study the extension of SAM's capabilities to the sequence of audio-visual scenes by analyzing contextual cross-modal relationships across the frames. To achieve this, we propose a Spatio-Temporal, Bidirectional Audio-Visual Attention (ST-BAVA) module integrated into the middle of SAM's image encoder and mask decoder. It adaptively updates the audio-visual features to convey the spatio-temporal correspondence between the video frames and audio streams. Extensive experiments demonstrate that our proposed model outperforms the state-of-the-art methods on AVS benchmarks, especially with an 8.3% mIoU gain on a challenging multi-sources subset.

Changho Jo, Woobin Im, Sung-Eui Yoon

In computer vision, recovering spatial information by filling in masked regions, e.g., inpainting, has been widely investigated for its usability and wide applicability to other various applications: image inpainting, image extrapolation, and environment map estimation. Most of them are studied separately depending on the applications. Our focus, however, is on accommodating the opposite task, e.g., image outpainting, which would benefit the target applications, e.g., image inpainting. Our self-supervision method, In-N-Out, is summarized as a training approach that leverages the knowledge of the opposite task into the target model. We empirically show that In-N-Out -- which explores the complementary information -- effectively takes advantage over the traditional pipelines where only task-specific learning takes place in training. In experiments, we compare our method to the traditional procedure and analyze the effectiveness of our method on different applications: image inpainting, image extrapolation, and environment map estimation. For these tasks, we demonstrate that In-N-Out consistently improves the performance of the recent works with In-N-Out self-supervision to their training procedure. Also, we show that our approach achieves better results than an existing training approach for outpainting.

Moonyoung Lee, Youngsun Kwon, Sebin Lee et al.

Although bipedal locomotion provides the ability to traverse unstructured environments, it requires careful planning and control to safely walk across without falling. This poses an integrated challenge for the robot to perceive, plan, and control its movements, especially with dynamic motions where the robot may have to adapt its swing-leg trajectory onthe-fly in order to safely place its foot on the uneven terrain. In this paper we present an efficient geometric footstep planner and the corresponding walking controller that enables a humanoid robot to dynamically walk across uneven terrain at speeds up to 0.3 m/s. As dynamic locomotion, we refer first to the continuous walking motion without stopping, and second to the on-the-fly replanning of the landing footstep position in middle of the swing phase during the robot gait cycle. This is mainly achieved through the streamlined integration between an efficient sampling-based planner and robust walking controller. The footstep planner is able to generate feasible footsteps within 5 milliseconds, and the controller is able to generate a new corresponding swing leg trajectory as well as the wholebody motion to dynamically balance the robot to the newly updated footsteps. The proposed perception-control pipeline is evaluated and demonstrated with real experiments using a fullscale humanoid to traverse across uneven terrains featured by static stepping stones, dynamically movable stepping stone, or narrow path.

Mingi Lim, Sung-eui Yoon

The process of decomposing target images into their internal properties is a difficult task due to the inherent ill-posed nature of the problem. The lack of data required to train a network is a one of the reasons why the decomposing appearance task is difficult. In this paper, we propose a deep learning-based reflectance map prediction system for material estimation of target objects in the image, so as to alleviate the ill-posed problem that occurs in this image decomposition operation. We also propose a network architecture for Bidirectional Reflectance Distribution Function (BRDF) parameter estimation, environment map estimation. We also use synthetic data to solve the lack of data problems. We get out of the previously proposed Deep Learning-based network architecture for reflectance map, and we newly propose to use conditional Generative Adversarial Network (cGAN) structures for estimating the reflectance map, which enables better results in many applications. To improve the efficiency of learning in this structure, we newly utilized the loss function using the normal map of the target object.

Seongjae Kang, Jaeyoon Kim, Sung-eui Yoon

In this research, Piano performances have been analyzed only based on visual information. Computer vision algorithms, e.g., Hough transform and binary thresholding, have been applied to find where the keyboard and specific keys are located. At the same time, Convolutional Neural Networks(CNNs) has been also utilized to find whether specific keys are pressed or not, and how much intensity the keys are pressed only based on visual information. Especially for detecting intensity, a new method of utilizing spatial, temporal CNNs model is devised. Early fusion technique is especially applied in temporal CNNs architecture to analyze hand movement. We also make a new dataset for training each model. Especially when finding an intensity of a pressed key, both of video frames and their optical flow images are used to train models to find effectiveness.

Mincheul Kang, Heechan Shin, Donghyuk Kim et al.

A redundant manipulator has multiple inverse kinematics solutions per end-effector pose. Accordingly, there can be many trajectories for joints that follow a given endeffector path in the Cartesian space. In this paper, we present a trajectory optimization of a redundant manipulator (TORM) to synthesize a trajectory that follows a given end-effector path accurately, while achieving smoothness and collisionfree manipulation. Our method holistically incorporates three desired properties into the trajectory optimization process by integrating the Jacobian-based inverse kinematics solving method and an optimization-based motion planning approach. Specifically, we optimize a trajectory using two-stage gradient descent to reduce potential competition between different properties during the update. To avoid falling into local minima, we iteratively explore different candidate trajectories with our local update. We compare our method with state-of-the-art methods in test scenes including external obstacles and two non-obstacle problems. Our method robustly minimizes the pose error in a progressive manner while satisfying various desirable properties.

Taeyoung Kim, Youngsun Kwon, Sung-eui Yoon

This paper proposes a real-time system integrating an acoustic material estimation from visual appearance and an on-the-fly mapping in the 3-dimension. The proposed method estimates the acoustic materials of surroundings in indoor scenes and incorporates them to a 3-D occupancy map, as a robot moves around the environment. To estimate the acoustic material from the visual cue, we apply the state-of-the-art semantic segmentation CNN network based on the assumption that the visual appearance and the acoustic materials have a strong association. Furthermore, we introduce an update policy to handle the material estimations during the online mapping process. As a result, our environment map with acoustic material can be used for sound-related robotics applications, such as sound source localization taking into account various acoustic propagation (e.g., reflection).

Hongsun Choi, Mincheul Kang, Youngsun Kwon et al.

We propose a novel method utilizing an objectness score for maintaining the locations and classes of objects detected from Mask R-CNN during mobile robot navigation. The objectness score is defined to measure how well the detector identifies the locations and classes of objects during navigation. Specifically, it is designed to increase when there is sufficient distance between a detected object and the camera. During the navigation process, we transform the locations of objects in 3D world coordinates into 2D image coordinates through an affine projection and decide whether to retain the classes of detected objects using the objectness score. We conducted experiments to determine how well the locations and classes of detected objects are maintained at various angles and positions. Experimental results showed that our approach is efficient and robust, regardless of changing angles and distances.

Nitin Agarwal, Sung-eui Yoon, M Gopi

Sharp features such as edges and corners play an important role in the perception of 3D models. In order to capture them better, we propose quadric loss, a point-surface loss function, which minimizes the quadric error between the reconstructed points and the input surface. Computation of Quadric loss is easy, efficient since the quadric matrices can be computed apriori, and is fully differentiable, making quadric loss suitable for training point and mesh based architectures. Through extensive experiments we show the merits and demerits of quadric loss. When combined with Chamfer loss, quadric loss achieves better reconstruction results as compared to any one of them or other point-surface loss functions.

Chiwan Song, Woobin Im, Sung-eui Yoon

Understanding the content of videos is one of the core techniques for developing various helpful applications in the real world, such as recognizing various human actions for surveillance systems or customer behavior analysis in an autonomous shop. However, understanding the content or story of the video still remains a challenging problem due to its sheer amount of data and temporal structure. In this paper, we propose a multi-channel neural network structure that adopts a two-stream network structure, which has been shown high performance in human action recognition field, and use it as a spatiotemporal video feature extractor for solving video question and answering task. We also adopt a squeeze-and-excitation structure to two-stream network structure for achieving a channel-wise attended spatiotemporal feature. For jointly modeling the spatiotemporal features from video and the textual features from the question, we design a context matching module with a level adjusting layer to remove the gap of information between visual and textual features by applying attention mechanism on joint modeling. Finally, we adopt a scoring mechanism and smoothed ranking loss objective function for selecting the correct answer from answer candidates. We evaluate our model with TVQA dataset, and our approach shows the improved result in textual only setting, but the result with visual feature shows the limitation and possibility of our approach.

JaeWon Choi, Sung-eui Yoon

At an early age, human infants are able to learn and build a model of the world very quickly by constantly observing and interacting with objects around them. One of the most fundamental intuitions human infants acquire is intuitive physics. Human infants learn and develop these models, which later serve as prior knowledge for further learning. Inspired by such behaviors exhibited by human infants, we introduce a graphical physics network integrated with deep reinforcement learning. Specifically, we introduce an intrinsic reward normalization method that allows our agent to efficiently choose actions that can improve its intuitive physics model the most. Using a 3D physics engine, we show that our graphical physics network is able to infer object's positions and velocities very effectively, and our deep reinforcement learning network encourages an agent to improve its model by making it continuously interact with objects only using intrinsic motivation. We experiment our model in both stationary and non-stationary state problems and show benefits of our approach in terms of the number of different actions the agent performs and the accuracy of agent's intuition model. Videos are at https://www.youtube.com/watch?v=pDbByp91r3M&t=2s

Soomin Kim, Yuchi Huo, Sung-Eui Yoon

Recently, deep learning-based single image reflection separation methods have been exploited widely. To benefit the learning approach, a large number of training image pairs (i.e., with and without reflections) were synthesized in various ways, yet they are away from a physically-based direction. In this paper, physically based rendering is used for faithfully synthesizing the required training images, and a corresponding network structure and loss term are proposed. We utilize existing RGBD/RGB images to estimate meshes, then physically simulate the light transportation between meshes, glass, and lens with path tracing to synthesize training data, which successfully reproduce the spatially variant anisotropic visual effect of glass reflection. For guiding the separation better, we additionally consider a module, backtrack network (BT-net) for backtracking the reflections, which removes complicated ghosting, attenuation, blurred and defocused effect of glass/lens. This enables obtaining a priori information before having the distortion. The proposed method considering additional a priori information with physically simulated training data is validated with various real reflection images and shows visually pleasant and numerical advantages compared with state-of-the-art techniques.

Inkyu An, Doheon Lee, Byeongho Jo et al.

We present a novel, robust sound source localization algorithm considering back-propagation signals. Sound propagation paths are estimated by generating direct and reflection acoustic rays based on ray tracing in a backward manner. We then compute the back-propagation signals by designing and using the impulse response of the backward sound propagation based on the acoustic ray paths. For identifying the 3D source position, we suggest a localization method based on the Monte Carlo localization algorithm. Candidates for a source position is determined by identifying the convergence regions of acoustic ray paths. This candidate is validated by measuring similarities between back-propagation signals, under the assumption that the back-propagation signals of different acoustic ray paths should be similar near the sound source position. Thanks to considering similarities of back-propagation signals, our approach can localize a source position with an averaged error of 0.51 m in a room of 7 m by 7 m area with 3 m height in tested environments. We also observe 65 % to 220 % improvement in accuracy over the stateof-the-art method. This improvement is achieved in environments containing a moving source, an obstacle, and noises.

Inkyu An, Doheon Lee, Jung-woo Choi et al.

We present a novel sound localization algorithm for a non-line-of-sight (NLOS) sound source in indoor environments. Our approach exploits the diffraction properties of sound waves as they bend around a barrier or an obstacle in the scene. We combine a ray tracing based sound propagation algorithm with a Uniform Theory of Diffraction (UTD) model, which simulate bending effects by placing a virtual sound source on a wedge in the environment. We precompute the wedges of a reconstructed mesh of an indoor scene and use them to generate diffraction acoustic rays to localize the 3D position of the source. Our method identifies the convergence region of those generated acoustic rays as the estimated source position based on a particle filter. We have evaluated our algorithm in multiple scenarios consisting of a static and dynamic NLOS sound source. In our tested cases, our approach can localize a source position with an average accuracy error, 0.7m, measured by the L2 distance between estimated and actual source locations in a 7m*7m*3m room. Furthermore, we observe 37% to 130% improvement in accuracy over a state-of-the-art localization method that does not model diffraction effects, especially when a sound source is not visible to the robot.