arXiv subjects

Seunghwan Lee

Publications and source records attributed to Seunghwan Lee.

At least 19 recordsLinked to original sources

arXiv2026

MultihopSpatial: Multi-hop Compositional Spatial Reasoning Benchmark for Vision-Language Model

Spatial reasoning is foundational for Vision-Language Models (VLMs), particularly when deployed as Vision-Language-Action (VLA) agents in physical environments. However, existing benchmarks predominantly focus on elementary, single-hop relations, neglecting the multi-hop compositional reasoning and precise visual grounding essential for real-world scenarios. To address this, we introduce MultihopSpatial, offering three key contributions: (1) A comprehensive benchmark designed for multi-hop and compositional spatial reasoning, featuring 1- to 3-hop complex queries across diverse spatial perspectives. (2) Acc@50IoU, a complementary metric that simultaneously evaluates reasoning and visual grounding by requiring both answer selection and precise bounding box prediction - capabilities vital for robust VLA deployment. (3) MultihopSpatial-Train, a dedicated large-scale training corpus to foster spatial intelligence. Extensive evaluation of 37 state-of-the-art VLMs yields eight key insights, revealing that compositional spatial reasoning remains a formidable challenge. Finally, we demonstrate that reinforcement learning post-training on our corpus enhances both intrinsic VLM spatial reasoning and downstream embodied manipulation performance.

cs.CV↗

arXiv2026

Mechanical isotropy of heterogeneous octahedral materials

An octahedral network has been widely used as a fundamental building block for diverse architected materials. Here, we demonstrate that heterogeneous octahedral materials can achieve complete elastic isotropy at a critical constituent volume fraction, nearly independent of the constituent stiffness ratio. The anisotropy ratio, a, transitions from a > 1 to a < 1 at the critical constituent volume fraction, due to a change in the dominant deformation modes. Microstructural analysis reveals that the geometry and connectivity of the heterogeneous octahedral materials are very similar to those of heterogeneous materials constructed on combined simple cubic (SC) and body-centered cubic (BCC) lattices exhibiting opposite elastic anisotropy. More importantly, we demonstrate that varying the constituent volume fractions in the octahedral materials governs elastic anisotropy, similarly to tuning the BCC-to-SC composition ratio in the SC-BCC materials widely employed for designing mechanically isotropic architected materials. The mechanical isotropy of the heterogeneous octahedral materials is further assessed using 3D-printed prototypes.

cond-mat.soft↗

arXiv2026

SyMerge: From Non-Interference to Synergistic Merging via Single-Layer Adaptation

Model merging combines independently trained models into a single multi-task model. However, most existing approaches focus primarily on avoiding task interference. We argue that its greater potential lies in enabling task synergy, where tasks actively improve one another. We identify cross-task performance, defined by compatibility between encoders and predictors across tasks, as a key indicator of merge quality. We demonstrate that adapting only a single task-specific layer is sufficient to induce such synergy. This study proposes SyMerge, a lightweight framework that jointly optimizes merging coefficients and a single task-specific layer. We adopt an expert-guided self-labeling objective, providing stable supervision beyond entropy minimization. Intriguingly, we further show that SyMerge successfully merges models trained from different initializations, a regime where standard methods break down. Our minimalist yet principled method achieves state-of-the-art results across vision, dense prediction, and NLP benchmarks. Our code is available at https://aim-skku.github.io/SyMerge

cs.LG↗

arXiv2026

Mechanical anisotropy of 3D-printed digital materials at large strains

3D-printed digital materials whose mechanical behavior travels between those from thermoplastic to rubbery polymers have become increasingly important. However, their mechanical functionalities have not been fully exploited due to intrinsic mechanical anisotropy resulting from microstructural heterogeneity. Here, we combine mechanical testing, microscopy analysis and micromechanical modeling for a comprehensive understanding of complex deformation mechanisms responsible for the printing-orientation-dependent nonlinear mechanical behavior of digital materials at small to large strains. Towards this end, we construct representative volume elements that account for highly anisotropic microstructural features resulting from the printing-orientation-dependent diffusion and mixing between photocurable base resins. We then demonstrate, through micromechanical analysis, that stable compressive deformation of well-aligned elliptical hard thermoplastic inclusions embedded within the surrounding soft rubbery matrix gives rise to initial elastic anisotropy. Our experimental and micromechanical modeling results also show that the interplay between buckling instability and plastic deformation of the high-aspect-ratio hard domains governs mechanical anisotropy at large strains as well as the printing-orientation-dependent resilience and energy dissipation capabilities in these digital materials.

cond-mat.soft↗

arXiv2025

Enhancing Topological Dependencies in Spatio-Temporal Graphs with Cycle Message Passing Blocks

Graph Neural Networks (GNNs) and Transformer-based models have been increasingly adopted to learn the complex vector representations of spatio-temporal graphs, capturing intricate spatio-temporal dependencies crucial for applications such as traffic datasets. Although many existing methods utilize multi-head attention mechanisms and message-passing neural networks (MPNNs) to capture both spatial and temporal relations, these approaches encode temporal and spatial relations independently, and reflect the graph's topological characteristics in a limited manner. In this work, we introduce the Cycle to Mixer (Cy2Mixer), a novel spatio-temporal GNN based on topological non-trivial invariants of spatio-temporal graphs with gated multi-layer perceptrons (gMLP). The Cy2Mixer is composed of three blocks based on MLPs: A temporal block for capturing temporal properties, a message-passing block for encapsulating spatial information, and a cycle message-passing block for enriching topological information through cyclic subgraphs. We bolster the effectiveness of Cy2Mixer with mathematical evidence emphasizing that our cycle message-passing block is capable of offering differentiated information to the deep learning model compared to the message-passing block. Furthermore, empirical evaluations substantiate the efficacy of the Cy2Mixer, demonstrating state-of-the-art performances across various spatio-temporal benchmark datasets. The source code is available at https://github.com/leemingo/cy2mixer.

cs.LG↗

arXiv2025

Instance-Aware Test-Time Segmentation for Continual Domain Shifts

Continual Test-Time Adaptation (CTTA) enables pre-trained models to adapt to continuously evolving domains. Existing methods have improved robustness but typically rely on fixed or batch-level thresholds, which cannot account for varying difficulty across classes and instances. This limitation is especially problematic in semantic segmentation, where each image requires dense, multi-class predictions. We propose an approach that adaptively adjusts pseudo labels to reflect the confidence distribution within each image and dynamically balances learning toward classes most affected by domain shifts. This fine-grained, class- and instance-aware adaptation produces more reliable supervision and mitigates error accumulation throughout continual adaptation. Extensive experiments across eight CTTA and TTA scenarios, including synthetic-to-real and long-term shifts, show that our method consistently outperforms state-of-the-art techniques, setting a new standard for semantic segmentation under evolving conditions.

cs.CV↗

arXiv2025

Hi-SAFE: Hierarchical Secure Aggregation for Lightweight Federated Learning

Federated learning (FL) faces challenges in ensuring both privacy and communication efficiency, particularly in resource-constrained environments such as Internet of Things (IoT) and edge networks. While sign-based methods, such as sign stochastic gradient descent with majority voting (SIGNSGD-MV), offer substantial bandwidth savings, they remain vulnerable to inference attacks due to exposure of gradient signs. Existing secure aggregation techniques are either incompatible with sign-based methods or incur prohibitive overhead. To address these limitations, we propose Hi-SAFE, a lightweight and cryptographically secure aggregation framework for sign-based FL. Our core contribution is the construction of efficient majority vote polynomials for SIGNSGD-MV, derived from Fermat's Little Theorem. This formulation represents the majority vote as a low-degree polynomial over a finite field, enabling secure evaluation that hides intermediate values and reveals only the final result. We further introduce a hierarchical subgrouping strategy that ensures constant multiplicative depth and bounded per-user complexity, independent of the number of users n.

cs.LG↗

arXiv2025

Task Vector Quantization for Memory-Efficient Model Merging

Model merging enables efficient multi-task models by combining task-specific fine-tuned checkpoints. However, storing multiple task-specific checkpoints requires significant memory, limiting scalability and restricting model merging to larger models and diverse tasks. In this paper, we propose quantizing task vectors (i.e., the difference between pre-trained and fine-tuned checkpoints) instead of quantizing fine-tuned checkpoints. We observe that task vectors exhibit a narrow weight range, enabling low precision quantization (e.g., 4 bit) within existing task vector merging frameworks. To further mitigate quantization errors within ultra-low bit precision (e.g., 2 bit), we introduce Residual Task Vector Quantization, which decomposes the task vector into a base vector and offset component. We allocate bits based on quantization sensitivity, ensuring precision while minimizing error within a memory budget. Experiments on image classification and dense prediction show our method maintains or improves model merging performance while using only 8% of the memory required for full-precision checkpoints.

cs.LG↗

arXiv2025

Stiffness, strength, energy dissipation and reusability in heterogeneous architected polycrystals

We design, fabricate and test heterogeneous architected polycrystals, composed of hard plastomers and soft elastomers, which thus show outstanding mechanical resilience and energy dissipation simultaneously. Grain boundaries that separate randomly oriented single crystalline grains is carefully designed, first enabling coherent connectivity and strength in the grain boundary regions throughout the polycrystalline network. By combining experiments and numerical simulations on 3D-printed prototypes, we show that the interplay between grain interiors and grain boundaries is responsible for the grain-size effects emerging in these architected materials, analogous to those in their atomic or metallic counterparts. Furthermore, direct visualization of inter- and intra-grain deformation and failure mechanisms at the macroscopic scale reveals that crystallographic texture throughout the polycrystalline aggregates plays a fundamental role in the key mechanical features in our new heterogeneous polycrystals. Our results show that the engineered grain boundary and crystallographic texture not only modify the highly resilient yet dissipative global responses but also critically influence reusability in this new class of architected materials.

cond-mat.soft↗

arXiv2024

InFusionSurf: Refining Neural RGB-D Surface Reconstruction Using Per-Frame Intrinsic Refinement and TSDF Fusion Prior Learning

We introduce InFusionSurf, an innovative enhancement for neural radiance field (NeRF) frameworks in 3D surface reconstruction using RGB-D video frames. Building upon previous methods that have employed feature encoding to improve optimization speed, we further improve the reconstruction quality with minimal impact on optimization time by refining depth information. InFusionSurf addresses camera motion-induced blurs in each depth frame through a per-frame intrinsic refinement scheme. It incorporates the truncated signed distance field (TSDF) Fusion, a classical real-time 3D surface reconstruction method, as a pretraining tool for the feature grid, enhancing reconstruction details and training speed. Comparative quantitative and qualitative analyses show that InFusionSurf reconstructs scenes with high accuracy while maintaining optimization efficiency. The effectiveness of our intrinsic refinement and TSDF Fusion-based pretraining is further validated through an ablation study.

cs.CV↗

arXiv2024

DiffusionPoser: Real-time Human Motion Reconstruction From Arbitrary Sparse Sensors Using Autoregressive Diffusion

Motion capture from a limited number of body-worn sensors, such as inertial measurement units (IMUs) and pressure insoles, has important applications in health, human performance, and entertainment. Recent work has focused on accurately reconstructing whole-body motion from a specific sensor configuration using six IMUs. While a common goal across applications is to use the minimal number of sensors to achieve required accuracy, the optimal arrangement of the sensors might differ from application to application. We propose a single diffusion model, DiffusionPoser, which reconstructs human motion in real-time from an arbitrary combination of sensors, including IMUs placed at specified locations, and, pressure insoles. Unlike existing methods, our model grants users the flexibility to determine the number and arrangement of sensors tailored to the specific activity of interest, without the need for retraining. A novel autoregressive inferencing scheme ensures real-time motion reconstruction that closely aligns with measured sensor signals. The generative nature of DiffusionPoser ensures realistic behavior, even for degrees-of-freedom not directly measured. Qualitative results can be found on our website: https://diffusionposer.github.io/.

cs.CV↗

arXiv2023

Anatomically Detailed Simulation of Human Torso

Existing digital human models approximate the human skeletal system using rigid bodies connected by rotational joints. While the simplification is considered acceptable for legs and arms, it significantly lacks fidelity to model rich torso movements in common activities such as dancing, Yoga, and various sports. Research from biomechanics provides more detailed modeling for parts of the torso, but their models often operate in isolation and are not fast and robust enough to support computationally heavy applications and large-scale data generation for full-body digital humans. This paper proposes a new torso model that aims to achieve high fidelity both in perception and in functionality, while being computationally feasible for simulation and optimal control tasks. We build a detailed human torso model consisting of various anatomical components, including facets, ligaments, and intervertebral discs, by coupling efficient finite-element and rigid-body simulations. Given an existing motion capture sequence without dense markers placed on the torso, our new model is able to recover the underlying torso bone movements. Our method is remarkably robust that it can be used to automatically "retrofit" the entire Mixamo motion database of highly diverse human motions without user intervention. We also show that our model is computationally efficient for solving trajectory optimization of highly dynamic full-body movements, without relying on any reference motion. Physiological validity of the model is validated against established literature.

cs.GR↗

arXiv2023

NoiseTransfer: Image Noise Generation with Contrastive Embeddings

Deep image denoising networks have achieved impressive success with the help of a considerably large number of synthetic train datasets. However, real-world denoising is a still challenging problem due to the dissimilarity between distributions of real and synthetic noisy datasets. Although several real-world noisy datasets have been presented, the number of train datasets (i.e., pairs of clean and real noisy images) is limited, and acquiring more real noise datasets is laborious and expensive. To mitigate this problem, numerous attempts to simulate real noise models using generative models have been studied. Nevertheless, previous works had to train multiple networks to handle multiple different noise distributions. By contrast, we propose a new generative model that can synthesize noisy images with multiple different noise distributions. Specifically, we adopt recent contrastive learning to learn distinguishable latent features of the noise. Moreover, our model can generate new noisy images by transferring the noise characteristics solely from a single reference noisy image. We demonstrate the accuracy and the effectiveness of our noise model for both known and unknown noise removal.

cs.CV↗

arXiv2022

An efficient GPU-Parallel Coordinate Descent Algorithm for Sparse Precision Matrix Estimation via Scaled Lasso

The sparse precision matrix plays an essential role in the Gaussian graphical model since a zero off-diagonal element indicates conditional independence of the corresponding two variables given others. In the Gaussian graphical model, many methods have been proposed, and their theoretical properties are given as well. Among these, the sparse precision matrix estimation via scaled lasso (SPMESL) has an attractive feature in which the penalty level is automatically set to achieve the optimal convergence rate under the sparsity and invertibility conditions. Conversely, other methods need to be used in searching for the optimal tuning parameter. Despite such an advantage, the SPMESL has not been widely used due to its expensive computational cost. In this paper, we develop a GPU-parallel coordinate descent (CD) algorithm for the SPMESL and numerically show that the proposed algorithm is much faster than the least angle regression (LARS) tailored to the SPMESL. Several comprehensive numerical studies are conducted to investigate the scalability of the proposed algorithm and the estimation performance of the SPMESL. The results show that the SPMESL has the lowest false discovery rate for all cases and the best performance in the case where the level of the sparsity of the columns is high.

stat.CO↗

arXiv2022

Generating 3D Bio-Printable Patches Using Wound Segmentation and Reconstruction to Treat Diabetic Foot Ulcers

We introduce AiD Regen, a novel system that generates 3D wound models combining 2D semantic segmentation with 3D reconstruction so that they can be printed via 3D bio-printers during the surgery to treat diabetic foot ulcers (DFUs). AiD Regen seamlessly binds the full pipeline, which includes RGB-D image capturing, semantic segmentation, boundary-guided point-cloud processing, 3D model reconstruction, and 3D printable G-code generation, into a single system that can be used out of the box. We developed a multi-stage data preprocessing method to handle small and unbalanced DFU image datasets. AiD Regen's human-in-the-loop machine learning interface enables clinicians to not only create 3D regenerative patches with just a few touch interactions but also customize and confirm wound boundaries. As evidenced by our experiments, our model outperforms prior wound segmentation models and our reconstruction algorithm is capable of generating 3D wound models with compelling accuracy. We further conducted a case study on a real DFU patient and demonstrated the effectiveness of AiD Regen in treating DFU wounds.

eess.IV↗

arXiv2021

Cycled Compositional Learning between Images and Text

We present an approach named the Cycled Composition Network that can measure the semantic distance of the composition of image-text embedding. First, the Composition Network transit a reference image to target image in an embedding space using relative caption. Second, the Correction Network calculates a difference between reference and retrieved target images in the embedding space and match it with a relative caption. Our goal is to learn a Composition mapping with the Composition Network. Since this one-way mapping is highly under-constrained, we couple it with an inverse relation learning with the Correction Network and introduce a cycled relation for given Image We participate in Fashion IQ 2020 challenge and have won the first place with the ensemble of our model.

cs.CV↗

arXiv2021

An efficient parallel block coordinate descent algorithm for large-scale precision matrix estimation using graphics processing units

Large-scale sparse precision matrix estimation has attracted wide interest from the statistics community. The convex partial correlation selection method (CONCORD) developed by Khare et al. (2015) has recently been credited with some theoretical properties for estimating sparse precision matrices. The CONCORD obtains its solution by a coordinate descent algorithm (CONCORD-CD) based on the convexity of the objective function. However, since a coordinate-wise update in CONCORD-CD is inherently serial, a scale-up is nontrivial. In this paper, we propose a novel parallelization of CONCORD-CD, namely, CONCORD-PCD. CONCORD-PCD partitions the off-diagonal elements into several groups and updates each group simultaneously without harming the computational convergence of CONCORD-CD. We guarantee this by employing the notion of edge coloring in graph theory. Specifically, we establish a nontrivial correspondence between scheduling the updates of the off-diagonal elements in CONCORD-CD and coloring the edges of a complete graph. It turns out that CONCORD-PCD simultaneously updates off-diagonal elements in which the associated edges are colorable with the same color. As a result, the number of steps required for updating off-diagonal elements reduces from p(p-1)/2 to p-1 (for even p) or p (for odd p), where p denotes the number of variables. We prove that the number of such steps is irreducible In addition, CONCORD-PCD is tailored to single-instruction multiple-data (SIMD) parallelism. A numerical study shows that the SIMD-parallelized PCD algorithm implemented in graphics processing units (GPUs) boosts the CONCORD-CD algorithm multiple times.

stat.CO↗

arXiv2021

Restore from Restored: Video Restoration with Pseudo Clean Video

In this study, we propose a self-supervised video denoising method called "restore-from-restored." This method fine-tunes a pre-trained network by using a pseudo clean video during the test phase. The pseudo clean video is obtained by applying a noisy video to the baseline network. By adopting a fully convolutional neural network (FCN) as the baseline, we can improve video denoising performance without accurate optical flow estimation and registration steps, in contrast to many conventional video restoration methods, due to the translation equivariant property of the FCN. Specifically, the proposed method can take advantage of plentiful similar patches existing across multiple consecutive frames (i.e., patch-recurrence); these patches can boost the performance of the baseline network by a large margin. We analyze the restoration performance of the fine-tuned video denoising networks with the proposed self-supervision-based learning algorithm, and demonstrate that the FCN can utilize recurring patches without requiring accurate registration among adjacent frames. In our experiments, we apply the proposed method to state-of-the-art denoisers and show that our fine-tuned networks achieve a considerable improvement in denoising performance.

cs.CV↗