arXiv subjects

Junseok Kwon

Publications and source records attributed to Junseok Kwon.

17 recordsLinked to original sources

arXiv2026

URHead: A Unified UV-Space Representation for Joint Mesh-3DGS Optimization in Head Avatars

We present URHead, a unified representation for high-fidelity and animatable head avatars that fundamentally redefines mesh-Gaussian integration. While mesh-based methods offer precise geometric control but lack photorealistic detail, and Gaussian-based approaches achieve photorealism but suffer from poor structural consistency, existing hybrid solutions fail to fully leverage their complementary strengths. Our key contribution is a UV-space unification where both representations share a common UV parameterization. Through joint optimization with adaptive gaussian sampling, our method automatically learns to disentangle and allocate appropriate roles to each component. URHead maintains full parametric controllability while preserving subject-specific details, and outperforms existing state-of-the-art methods in reconstruction quality and animation consistency.

cs.GR↗

arXiv2026

$C^3$ASD: Multi-Level Consistency-Driven Representation Learning

Active Speaker Detection determines whether a visible person in a video is speaking at each moment. While recent audio-visual fusion methods perform well on clean data, they degrade under real-world corruptions such as background noise, occlusion, or simultaneous modality degradation. We attribute this limitation to the absence of explicit consistency constraints that promote robust, semantically aligned representations across modalities. Without such guidance, models tend to learn fragile modality-specific shortcuts that fail under corrupted conditions. We propose $C^3$ASD, a multi-level consistency-driven framework with three complementary constraints: embedding-level inter-modality consistency aligns audio-visual representations during speech; sequence-level intra-modality consistency separates speaking and non-speaking clusters via track-aware contrastive learning; and prediction-level consistency stabilizes fusion through knowledge distillation. Extensive experiments demonstrate significant improvements under diverse audio, visual and joint corruptions, while maintaining competitive performance on clean data.

cs.CV↗

arXiv2025

Lightweight Wasserstein Audio-Visual Model for Unified Speech Enhancement and Separation

Speech Enhancement (SE) and Speech Separation (SS) have traditionally been treated as distinct tasks in speech processing. However, real-world audio often involves both background noise and overlapping speakers, motivating the need for a unified solution. While recent approaches have attempted to integrate SE and SS within multi-stage architectures, these approaches typically involve complex, parameter-heavy models and rely on supervised training, limiting scalability and generalization. In this work, we propose UniVoiceLite, a lightweight and unsupervised audio-visual framework that unifies SE and SS within a single model. UniVoiceLite leverages lip motion and facial identity cues to guide speech extraction and employs Wasserstein distance regularization to stabilize the latent space without requiring paired noisy-clean data. Experimental results demonstrate that UniVoiceLite achieves strong performance in both noisy and multi-speaker scenarios, combining efficiency with robust generalization. The source code is available at https://github.com/jisoo-o/UniVoiceLite.

cs.CV↗

arXiv2025

Activation Boundary Matching: Task-Informed Initialization for Low-Rank Adaptation

Low-Rank Adaptation (LoRA) is highly sensitive to initialization, yet existing schemes construct the initial subspace from statistics at the pretrained point, capturing pre-adaptation geometry rather than how the adapter must move during learning. We examine the early adaptation trajectory and uncover a temporal asymmetry: task-induced activation boundaries---the signs of layer-wise pre-activations---recover markedly faster than activation values or effective low-rank updates and become reusable well before the probe adapter converges. Motivated by this, we propose Activation Boundary Matching for LoRA (ABM-LoRA), which briefly trains a standard adapter as an early task probe, then uses its pre-activation signs as layer-wise targets for a fresh adapter under a margin-based hinge objective. Both are discarded before otherwise unchanged downstream fine-tuning. We show that this boundary-supervision lifting can expose LoRA directions attenuated or locally unobservable through the downstream Jacobian, and that what transfers is the boundary side---not the exact activation value or the output-level signal. Because a partial boundary trace suffices, ABM-LoRA recovers most of the benefit of a fully trained reference at a fraction of the overhead, requiring only probe forward passes. It improves over standard LoRA on T5-base/GLUE, ConvNeXt-T and Swin-T fine-grained classification, and instruction tuning with Qwen2.5-1.5B and LLaMA2-7B, and matches or surpasses SVD- and gradient-based initializers without their preprocessing.

cs.CV↗

arXiv2025

Real-Aware Residual Model Merging for Deepfake Detection

Deepfake generators evolve quickly, making exhaustive data collection and repeated retraining impractical. We argue that model merging is a natural fit for deepfake detection: unlike generic multi-task settings with disjoint labels, deepfake specialists share the same binary decision and differ in generator-specific artifacts. Empirically, we show that simple weight averaging preserves Real representations while attenuating Fake-specific cues. Building upon these findings, we propose Real-aware Residual Model Merging (R$^2$M), a training-free parameter-space merging framework. R$^2$M estimates a shared Real component via a low-rank factorization of task vectors, decomposes each specialist into a Real-aligned part and a Fake residual, denoises residuals with layerwise rank truncation, and aggregates them with per-task norm matching to prevent any single generator from dominating. A concise rationale explains why a simple head suffices: the Real component induces a common separation direction in feature space, while truncated residuals contribute only minor off-axis variations. Across in-distribution, cross-dataset, and unseen-dataset, R$^2$M outperforms joint training and other merging baselines. Importantly, R$^2$M is also composable: when a new forgery family appears, we fine-tune one specialist and re-merge, eliminating the need for retraining.

cs.CV↗

arXiv2024

Uncertainty Calibration with Energy Based Instance-wise Scaling in the Wild Dataset

With the rapid advancement in the performance of deep neural networks (DNNs), there has been significant interest in deploying and incorporating artificial intelligence (AI) systems into real-world scenarios. However, many DNNs lack the ability to represent uncertainty, often exhibiting excessive confidence even when making incorrect predictions. To ensure the reliability of AI systems, particularly in safety-critical cases, DNNs should transparently reflect the uncertainty in their predictions. In this paper, we investigate robust post-hoc uncertainty calibration methods for DNNs within the context of multi-class classification tasks. While previous studies have made notable progress, they still face challenges in achieving robust calibration, particularly in scenarios involving out-of-distribution (OOD). We identify that previous methods lack adaptability to individual input data and struggle to accurately estimate uncertainty when processing inputs drawn from the wild dataset. To address this issue, we introduce a novel instance-wise calibration method based on an energy model. Our method incorporates energy scores instead of softmax confidence scores, allowing for adaptive consideration of DNN uncertainty for each prediction within a logit space. In experiments, we show that the proposed method consistently maintains robust performance across the spectrum, spanning from in-distribution to OOD scenarios, when compared to other state-of-the-art methods.

cs.LG↗

arXiv2024

Upright adjustment with graph convolutional networks

We present a novel method for the upright adjustment of 360 images. Our network consists of two modules, which are a convolutional neural network (CNN) and a graph convolutional network (GCN). The input 360 images is processed with the CNN for visual feature extraction, and the extracted feature map is converted into a graph that finds a spherical representation of the input. We also introduce a novel loss function to address the issue of discrete probability distributions defined on the surface of a sphere. Experimental results demonstrate that our method outperforms fully connected based methods.

cs.CV↗

arXiv2022

Spherical Transformer

Using convolutional neural networks for 360images can induce sub-optimal performance due to distortions entailed by a planar projection. The distortion gets deteriorated when a rotation is applied to the 360image. Thus, many researches based on convolutions attempt to reduce the distortions to learn accurate representation. In contrast, we leverage the transformer architecture to solve image classification problems for 360images. Using the proposed transformer for 360images has two advantages. First, our method does not require the erroneous planar projection process by sampling pixels from the sphere surface. Second, our sampling method based on regular polyhedrons makes low rotation equivariance errors, because specific rotations can be reduced to permutations of faces. In experiments, we validate our network on two aspects, as follows. First, we show that using a transformer with highly uniform sampling methods can help reduce the distortion. Second, we demonstrate that the transformer architecture can achieve rotation equivariance on specific rotations. We compare our method to other state-of-the-art algorithms using the SPH-MNIST, SPH-CIFAR, and SUN360 datasets and show that our method is competitive with other methods.

cs.CV↗

arXiv2021

Style Transfer with Target Feature Palette and Attention Coloring

Style transfer has attracted a lot of attentions, as it can change a given image into one with splendid artistic styles while preserving the image structure. However, conventional approaches easily lose image details and tend to produce unpleasant artifacts during style transfer. In this paper, to solve these problems, a novel artistic stylization method with target feature palettes is proposed, which can transfer key features accurately. Specifically, our method contains two modules, namely feature palette composition (FPC) and attention coloring (AC) modules. The FPC module captures representative features based on K-means clustering and produces a feature target palette. The following AC module calculates attention maps between content and style images, and transfers colors and patterns based on the attention map and the target palette. These modules enable the proposed stylization to focus on key features and generate plausibly transferred images. Thus, the contributions of the proposed method are to propose a novel deep learning-based style transfer method and present target feature palette and attention coloring modules, and provide in-depth analysis and insight on the proposed method via exhaustive ablation study. Qualitative and quantitative results show that our stylized images exhibit state-of-the-art performance, with strength in preserving core structures and details of the content image.

cs.CV↗

arXiv2020

DALE : Dark Region-Aware Low-light Image Enhancement

In this paper, we present a novel low-light image enhancement method called dark region-aware low-light image enhancement (DALE), where dark regions are accurately recognized by the proposed visual attention module and their brightness are intensively enhanced. Our method can estimate the visual attention in an efficient manner using super-pixels without any complicated process. Thus, the method can preserve the color, tone, and brightness of original images and prevents normally illuminated areas of the images from being saturated and distorted. Experimental results show that our method accurately identifies dark regions via the proposed visual attention, and qualitatively and quantitatively outperforms state-of-the-art methods.

eess.IV↗

arXiv2020

Visual Tracking by TridentAlign and Context Embedding

Recent advances in Siamese network-based visual tracking methods have enabled high performance on numerous tracking benchmarks. However, extensive scale variations of the target object and distractor objects with similar categories have consistently posed challenges in visual tracking. To address these persisting issues, we propose novel TridentAlign and context embedding modules for Siamese network-based visual tracking methods. The TridentAlign module facilitates adaptability to extensive scale variations and large deformations of the target, where it pools the feature representation of the target object into multiple spatial dimensions to form a feature pyramid, which is then utilized in the region proposal stage. Meanwhile, context embedding module aims to discriminate the target from distractor objects by accounting for the global context information among objects. The context embedding module extracts and embeds the global context information of a given frame into a local feature representation such that the information can be utilized in the final classification stage. Experimental results obtained on multiple benchmark datasets show that the performance of the proposed tracker is comparable to that of state-of-the-art trackers, while the proposed tracker runs at real-time speed.

cs.CV↗

arXiv2019

Adaptive Regularization via Residual Smoothing in Deep Learning Optimization

We present an adaptive regularization algorithm that can be effectively applied to the optimization problem in deep learning framework. Our regularization algorithm aims to take into account the fitness of data to the current state of model in the determination of regularity to achieve better generalization. The degree of regularization at each element in the target space of the neural network architecture is determined based on the residual at each optimization iteration in an adaptive way. Our adaptive regularization algorithm is designed to apply a diffusion process driven by the heat equation with spatially varying diffusivity depending on the probability density function following a certain distribution of residual. Our data-driven regularity is imposed by adaptively smoothing a simplified objective function in which the explicit regularization term is omitted in an alternating manner between the evaluation of residual and the determination of the degree of its regularity. The effectiveness of our algorithm is empirically demonstrated by the numerical experiments in the application of image classification problems, indicating that our algorithm outperforms other commonly used optimization algorithms in terms of generalization using popular deep learning models and benchmark datasets.

cs.LG↗

arXiv2019

Deep Meta Learning for Real-Time Target-Aware Visual Tracking

In this paper, we propose a novel on-line visual tracking framework based on the Siamese matching network and meta-learner network, which run at real-time speeds. Conventional deep convolutional feature-based discriminative visual tracking algorithms require continuous re-training of classifiers or correlation filters, which involve solving complex optimization tasks to adapt to the new appearance of a target object. To alleviate this complex process, our proposed algorithm incorporates and utilizes a meta-learner network to provide the matching network with new appearance information of the target objects by adding target-aware feature space. The parameters for the target-specific feature space are provided instantly from a single forward-pass of the meta-learner network. By eliminating the necessity of continuously solving complex optimization tasks in the course of tracking, experimental results demonstrate that our algorithm performs at a real-time speed while maintaining competitive performance among other state-of-the-art tracking algorithms.

cs.CV↗

arXiv2019

Deep ensemble network with explicit complementary model for accuracy-balanced classification

The average accuracy is one of major evaluation metrics for classification systems, while the accuracy deviation is another important performance metric used to evaluate various deep neural networks. In this paper, we present a new ensemble-like fast deep neural network, Harmony, that can reduce the accuracy deviation among categories without degrading overall average accuracy. Harmony consists of three sub-models, namely, Target model, Complementary model, and Conductor model. In Harmony, an object is classified by using either Target model or Complementary model. Target model is a conventional classification network for general categories, while Complementary model is a classification network especially for weak categories that are inaccurately classified by Target model. Conductor model is used to select one of two models. Experimental results demonstrate that Harmony accurately classifies categories, while it reduces the accuracy deviation among the categories.

cs.CV↗

arXiv2019

LED2Net: Deep Illumination-aware Dehazing with Low-light and Detail Enhancement

We present a novel dehazing and low-light enhancement method based on an illumination map that is accurately estimated by a convolutional neural network (CNN). In this paper, the illumination map is used as a component for three different tasks, namely, atmospheric light estimation, transmission map estimation, and low-light enhancement. To train CNNs for dehazing and low-light enhancement simultaneously based on the retinex theory, we synthesize numerous low-light and hazy images from normal hazy images from the FADE data set. In addition, we further improve the network using detail enhancement. Experimental results demonstrate that our method surpasses recent state-of-theart algorithms quantitatively and qualitatively. In particular, our haze-free images present vivid colors and enhance visibility without a halo effect or color distortion.

cs.CV↗

arXiv2019

3D Point Cloud Generative Adversarial Network Based on Tree Structured Graph Convolutions

In this paper, we propose a novel generative adversarial network (GAN) for 3D point clouds generation, which is called tree-GAN. To achieve state-of-the-art performance for multi-class 3D point cloud generation, a tree-structured graph convolution network (TreeGCN) is introduced as a generator for tree-GAN. Because TreeGCN performs graph convolutions within a tree, it can use ancestor information to boost the representation power for features. To evaluate GANs for 3D point clouds accurately, we develop a novel evaluation metric called Frechet point cloud distance (FPD). Experimental results demonstrate that the proposed tree-GAN outperforms state-of-the-art GANs in terms of both conventional metrics and FPD, and can generate point clouds for different semantic parts without prior knowledge.

cs.CV↗

arXiv2018

Real-time visual tracking by deep reinforced decision making

One of the major challenges of model-free visual tracking problem has been the difficulty originating from the unpredictable and drastic changes in the appearance of objects we target to track. Existing methods tackle this problem by updating the appearance model on-line in order to adapt to the changes in the appearance. Despite the success of these methods however, inaccurate and erroneous updates of the appearance model result in a tracker drift. In this paper, we introduce a novel real-time visual tracking algorithm based on a template selection strategy constructed by deep reinforcement learning methods. The tracking algorithm utilizes this strategy to choose the appropriate template for tracking a given frame. The template selection strategy is self-learned by utilizing a simple policy gradient method on numerous training episodes randomly generated from a tracking benchmark dataset. Our proposed reinforcement learning framework is generally applicable to other confidence map based tracking algorithms. The experiment shows that our tracking algorithm runs in real-time speed of 43 fps and the proposed policy network effectively decides the appropriate template for successful visual tracking.

cs.CV↗