arXiv subjects

Nojun Kwak

Publications and source records attributed to Nojun Kwak.

At least 55 records · Page 3Linked to original sources

arXiv2025

Understanding Differential Transformer Unchains Pretrained Self-Attentions

Differential Transformer has recently gained significant attention for its impressive empirical performance, often attributed to its ability to perform noise canceled attention. However, precisely how differential attention achieves its empirical benefits remains poorly understood. Moreover, Differential Transformer architecture demands large-scale training from scratch, hindering utilization of open pretrained weights. In this work, we conduct an in-depth investigation of Differential Transformer, uncovering three key factors behind its success: (1) enhanced expressivity via negative attention, (2) reduced redundancy among attention heads, and (3) improved learning dynamics. Based on these findings, we propose DEX, a novel method to efficiently integrate the advantages of differential attention into pretrained language models. By reusing the softmax attention scores and adding a lightweight differential operation on the output value matrix, DEX effectively incorporates the key advantages of differential attention while remaining lightweight in both training and inference. Evaluations confirm that DEX substantially improves the pretrained LLMs across diverse benchmarks, achieving significant performance gains with minimal adaptation data (< 0.01%).

cs.LG↗

arXiv2025

Generative Head-Mounted Camera Captures for Photorealistic Avatars

Enabling photorealistic avatar animations in virtual and augmented reality (VR/AR) has been challenging because of the difficulty of obtaining ground truth state of faces. It is physically impossible to obtain synchronized images from head-mounted cameras (HMC) sensing input, which has partial observations in infrared (IR), and an array of outside-in dome cameras, which have full observations that match avatars' appearance. Prior works relying on analysis-by-synthesis methods could generate accurate ground truth, but suffer from imperfect disentanglement between expression and style in their personalized training. The reliance of extensive paired captures (HMC and dome) for the same subject makes it operationally expensive to collect large-scale datasets, which cannot be reused for different HMC viewpoints and lighting. In this work, we propose a novel generative approach, Generative HMC (GenHMC), that leverages large unpaired HMC captures, which are much easier to collect, to directly generate high-quality synthetic HMC images given any conditioning avatar state from dome captures. We show that our method is able to properly disentangle the input conditioning signal that specifies facial expression and viewpoint, from facial appearance, leading to more accurate ground truth. Furthermore, our method can generalize to unseen identities, removing the reliance on the paired captures. We demonstrate these breakthroughs by both evaluating synthetic HMC images and universal face encoders trained from these new HMC-avatar correspondences, which achieve better data efficiency and state-of-the-art accuracy.

cs.CV↗

arXiv2025

Causal Interpretation of Sparse Autoencoder Features in Vision

Understanding what sparse auto-encoder (SAE) features in vision transformers truly represent is usually done by inspecting the patches where a feature's activation is highest. However, self-attention mixes information across the entire image, so an activated patch often co-occurs with-but does not cause-the feature's firing. We propose Causal Feature Explanation (CaFE), which leverages Effective Receptive Field (ERF). We consider each activation of an SAE feature to be a target and apply input-attribution methods to identify the image patches that causally drive that activation. Across CLIP-ViT features, ERF maps frequently diverge from naive activation maps, revealing hidden context dependencies (e.g., a "roaring face" feature that requires the co-occurrence of eyes and nose, rather than merely an open mouth). Patch insertion tests confirm that CaFE more effectively recovers or suppresses feature activations than activation-ranked patches. Our results show that CaFE yields more faithful and semantically precise explanations of vision-SAE features, highlighting the risk of misinterpretation when relying solely on activation location.

cs.CV↗

arXiv2025

The Role of Teacher Calibration in Knowledge Distillation

Knowledge Distillation (KD) has emerged as an effective model compression technique in deep learning, enabling the transfer of knowledge from a large teacher model to a compact student model. While KD has demonstrated significant success, it is not yet fully understood which factors contribute to improving the student's performance. In this paper, we reveal a strong correlation between the teacher's calibration error and the student's accuracy. Therefore, we claim that the calibration of the teacher model is an important factor for effective KD. Furthermore, we demonstrate that the performance of KD can be improved by simply employing a calibration method that reduces the teacher's calibration error. Our algorithm is versatile, demonstrating effectiveness across various tasks from classification to detection. Moreover, it can be easily integrated with existing state-of-the-art methods, consistently achieving superior performance.

cs.LG↗

arXiv2025

Real-Time Intuitive AI Drawing System for Collaboration: Enhancing Human Creativity through Formal and Contextual Intent Integration

This paper presents a real-time generative drawing system that interprets and integrates both formal intent - the structural, compositional, and stylistic attributes of a sketch - and contextual intent - the semantic and thematic meaning inferred from its visual content - into a unified transformation process. Unlike conventional text-prompt-based generative systems, which primarily capture high-level contextual descriptions, our approach simultaneously analyzes ground-level intuitive geometric features such as line trajectories, proportions, and spatial arrangement, and high-level semantic cues extracted via vision-language models. These dual intent signals are jointly conditioned in a multi-stage generation pipeline that combines contour-preserving structural control with style- and content-aware image synthesis. Implemented with a touchscreen-based interface and distributed inference architecture, the system achieves low-latency, two-stage transformation while supporting multi-user collaboration on shared canvases. The resulting platform enables participants, regardless of artistic expertise, to engage in synchronous, co-authored visual creation, redefining human-AI interaction as a process of co-creation and mutual enhancement.

cs.CV↗

arXiv2025

ROODI: Reconstructing Occluded Objects with Denoising Inpainters

While the quality of novel-view images has improved dramatically with 3D Gaussian Splatting, extracting specific objects from scenes remains challenging. Isolating individual 3D Gaussian primitives for each object and handling occlusions in scenes remains far from being solved. We propose a novel object extraction method based on two key principles: (1) object-centric reconstruction through removal of irrelevant primitives; and (2) leveraging generative inpainting to compensate for missing observations caused by occlusions. For pruning, we propose to remove irrelevant Gaussians by looking into how close they are to its K-nearest neighbors and removing those that are statistical outliers. Importantly, these distances must take into account the actual spatial extent they cover -- we thus propose to use Wasserstein distances. For inpainting, we employ an off-the-shelf diffusion-based inpainter combined with occlusion reasoning, utilizing the 3D representation of the entire scene. Our findings highlight the crucial synergy between proper pruning and inpainting, both of which significantly enhance extraction performance. We evaluate our method on a standard real-world dataset and introduce a synthetic dataset for quantitative analysis. Our approach outperforms the state-of-the-art, demonstrating its effectiveness in object extraction from complex scenes.

cs.CV↗

arXiv2025

Style Composition within Distinct LoRA modules for Traditional Art

Diffusion-based text-to-image models have achieved remarkable results in synthesizing diverse images from text prompts and can capture specific artistic styles via style personalization. However, their entangled latent space and lack of smooth interpolation make it difficult to apply distinct painting techniques in a controlled, regional manner, often causing one style to dominate. To overcome this, we propose a zero-shot diffusion pipeline that naturally blends multiple styles by performing style composition on the denoised latents predicted during the flow-matching denoising process of separately trained, style-specialized models. We leverage the fact that lower-noise latents carry stronger stylistic information and fuse them across heterogeneous diffusion pipelines using spatial masks, enabling precise, region-specific style control. This mechanism preserves the fidelity of each individual style while allowing user-guided mixing. Furthermore, to ensure structural coherence across different models, we incorporate depth-map conditioning via ControlNet into the diffusion framework. Qualitative and quantitative experiments demonstrate that our method successfully achieves region-specific style mixing according to the given masks.

cs.CV↗

arXiv2025

What's Making That Sound Right Now? Video-centric Audio-Visual Localization

Audio-Visual Localization (AVL) aims to identify sound-emitting sources within a visual scene. However, existing studies focus on image-level audio-visual associations, failing to capture temporal dynamics. Moreover, they assume simplified scenarios where sound sources are always visible and involve only a single object. To address these limitations, we propose AVATAR, a video-centric AVL benchmark that incorporates high-resolution temporal information. AVATAR introduces four distinct scenarios -- Single-sound, Mixed-sound, Multi-entity, and Off-screen -- enabling a more comprehensive evaluation of AVL models. Additionally, we present TAVLO, a novel video-centric AVL model that explicitly integrates temporal information. Experimental results show that conventional methods struggle to track temporal variations due to their reliance on global audio features and frame-level mappings. In contrast, TAVLO achieves robust and precise audio-visual alignment by leveraging high-resolution temporal modeling. Our work empirically demonstrates the importance of temporal dynamics in AVL and establishes a new standard for video-centric audio-visual localization.

cs.CV↗

arXiv2025

ReFlex: Text-Guided Editing of Real Images in Rectified Flow via Mid-Step Feature Extraction and Attention Adaptation

Rectified Flow text-to-image models surpass diffusion models in image quality and text alignment, but adapting ReFlow for real-image editing remains challenging. We propose a new real-image editing method for ReFlow by analyzing the intermediate representations of multimodal transformer blocks and identifying three key features. To extract these features from real images with sufficient structural preservation, we leverage mid-step latent, which is inverted only up to the mid-step. We then adapt attention during injection to improve editability and enhance alignment to the target text. Our method is training-free, requires no user-provided mask, and can be applied even without a source prompt. Extensive experiments on two benchmarks with nine baselines demonstrate its superior performance over prior methods, further validated by human evaluations confirming a strong user preference for our approach.

cs.CV↗

arXiv2025

Deep Support Vectors

Deep learning has achieved tremendous success. However, unlike SVMs, which provide direct decision criteria and can be trained with a small dataset, it still has significant weaknesses due to its requirement for massive datasets during training and the black-box characteristics on decision criteria. This paper addresses these issues by identifying support vectors in deep learning models. To this end, we propose the DeepKKT condition, an adaptation of the traditional Karush-Kuhn-Tucker (KKT) condition for deep learning models, and confirm that generated Deep Support Vectors (DSVs) using this condition exhibit properties similar to traditional support vectors. This allows us to apply our method to few-shot dataset distillation problems and alleviate the black-box characteristics of deep learning models. Additionally, we demonstrate that the DeepKKT condition can transform conventional classification models into generative models with high fidelity, particularly as latent generative models using class labels as latent variables. We validate the effectiveness of DSVs using common datasets (ImageNet, CIFAR10 and CIFAR100) on the general architectures (ResNet and ConvNet), proving their practical applicability.

cs.LG↗

arXiv2025

S3D: Sketch-Driven 3D Model Generation

Generating high-quality 3D models from 2D sketches is a challenging task due to the inherent ambiguity and sparsity of sketch data. In this paper, we present S3D, a novel framework that converts simple hand-drawn sketches into detailed 3D models. Our method utilizes a U-Net-based encoder-decoder architecture to convert sketches into face segmentation masks, which are then used to generate a 3D representation that can be rendered from novel views. To ensure robust consistency between the sketch domain and the 3D output, we introduce a novel style-alignment loss that aligns the U-Net bottleneck features with the initial encoder outputs of the 3D generation module, significantly enhancing reconstruction fidelity. To further enhance the network's robustness, we apply augmentation techniques to the sketch dataset. This streamlined framework demonstrates the effectiveness of S3D in generating high-quality 3D models from sketch inputs. The source code for this project is publicly available at https://github.com/hailsong/S3D.

cs.CV↗

arXiv2025

Unlocking the Potential of Unlabeled Data in Semi-Supervised Domain Generalization

We address the problem of semi-supervised domain generalization (SSDG), where the distributions of train and test data differ, and only a small amount of labeled data along with a larger amount of unlabeled data are available during training. Existing SSDG methods that leverage only the unlabeled samples for which the model's predictions are highly confident (confident-unlabeled samples), limit the full utilization of the available unlabeled data. To the best of our knowledge, we are the first to explore a method for incorporating the unconfident-unlabeled samples that were previously disregarded in SSDG setting. To this end, we propose UPCSC to utilize these unconfident-unlabeled samples in SSDG that consists of two modules: 1) Unlabeled Proxy-based Contrastive learning (UPC) module, treating unconfident-unlabeled samples as additional negative pairs and 2) Surrogate Class learning (SC) module, generating positive pairs for unconfident-unlabeled samples using their confusing class set. These modules are plug-and-play and do not require any domain labels, which can be easily integrated into existing approaches. Experiments on four widely used SSDG benchmarks demonstrate that our approach consistently improves performance when attached to baselines and outperforms competing plug-and-play methods. We also analyze the role of our method in SSDG, showing that it enhances class-level discriminability and mitigates domain gaps. The code is available at https://github.com/dongkwani/UPCSC.

cs.CV↗

arXiv2025

ARC-NeRF: Area Ray Casting for Broader Unseen View Coverage in Few-shot Object Rendering

Recent advancements in the Neural Radiance Field (NeRF) have enhanced its capabilities for novel view synthesis, yet its reliance on dense multi-view training images poses a practical challenge, often leading to artifacts and a lack of fine object details. Addressing this, we propose ARC-NeRF, an effective regularization-based approach with a novel Area Ray Casting strategy. While the previous ray augmentation methods are limited to covering only a single unseen view per extra ray, our proposed Area Ray covers a broader range of unseen views with just a single ray and enables an adaptive high-frequency regularization based on target pixel photo-consistency. Moreover, we propose luminance consistency regularization, which enhances the consistency of relative luminance between the original and Area Ray, leading to more accurate object textures. The relative luminance, as a free lunch extra data easily derived from RGB images, can be effectively utilized in few-shot scenarios where available training data is limited. Our ARC-NeRF outperforms its baseline and achieves competitive results on multiple benchmarks with sharply rendered fine details.

cs.CV↗

arXiv2025

A Revisit to the Decoder for Camouflaged Object Detection

Camouflaged object detection (COD) aims to generate a fine-grained segmentation map of camouflaged objects hidden in their background. Due to the hidden nature of camouflaged objects, it is essential for the decoder to be tailored to effectively extract proper features of camouflaged objects and extra-carefully generate their complex boundaries. In this paper, we propose a novel architecture that augments the prevalent decoding strategy in COD with Enrich Decoder and Retouch Decoder, which help to generate a fine-grained segmentation map. Specifically, the Enrich Decoder amplifies the channels of features that are important for COD using channel-wise attention. Retouch Decoder further refines the segmentation maps by spatially attending to important pixels, such as the boundary regions. With extensive experiments, we demonstrate that ENTO shows superior performance using various encoders, with the two novel components playing their unique roles that are mutually complementary.

cs.CV↗

arXiv2024

Respect the model: Fine-grained and Robust Explanation with Sharing Ratio Decomposition

The truthfulness of existing explanation methods in authentically elucidating the underlying model's decision-making process has been questioned. Existing methods have deviated from faithfully representing the model, thus susceptible to adversarial attacks. To address this, we propose a novel eXplainable AI (XAI) method called SRD (Sharing Ratio Decomposition), which sincerely reflects the model's inference process, resulting in significantly enhanced robustness in our explanations. Different from the conventional emphasis on the neuronal level, we adopt a vector perspective to consider the intricate nonlinear interactions between filters. We also introduce an interesting observation termed Activation-Pattern-Only Prediction (APOP), letting us emphasize the importance of inactive neurons and redefine relevance encapsulating all relevant information including both active and inactive neurons. Our method, SRD, allows for the recursive decomposition of a Pointwise Feature Vector (PFV), providing a high-resolution Effective Receptive Field (ERF) at any layer.

cs.CV↗

arXiv2024

Conservative Generator, Progressive Discriminator: Coordination of Adversaries in Few-shot Incremental Image Synthesis

The capacity to learn incrementally from an online stream of data is an envied trait of human learners, as deep neural networks typically suffer from catastrophic forgetting and stability-plasticity dilemma. Several works have previously explored incremental few-shot learning, a task with greater challenges due to data constraint, mostly in classification setting with mild success. In this work, we study the underrepresented task of generative incremental few-shot learning. To effectively handle the inherent challenges of incremental learning and few-shot learning, we propose a novel framework named ConPro that leverages the two-player nature of GANs. Specifically, we design a conservative generator that preserves past knowledge in parameter and compute efficient manner, and a progressive discriminator that learns to reason semantic distances between past and present task samples, minimizing overfitting with few data points and pursuing good forward transfer. We present experiments to validate the effectiveness of ConPro.

cs.CV↗

arXiv2024

Do not think about pink elephant!

Large Models (LMs) have heightened expectations for the potential of general AI as they are akin to human intelligence. This paper shows that recent large models such as Stable Diffusion and DALL-E3 also share the vulnerability of human intelligence, namely the "white bear phenomenon". We investigate the causes of the white bear phenomenon by analyzing their representation space. Based on this analysis, we propose a simple prompt-based attack method, which generates figures prohibited by the LM provider's policy. To counter these attacks, we introduce prompt-based defense strategies inspired by cognitive therapy techniques, successfully mitigating attacks by up to 48.22\%.

cs.CL↗

arXiv2024

MERLIN: Multimodal Embedding Refinement via LLM-based Iterative Navigation for Text-Video Retrieval-Rerank Pipeline

The rapid expansion of multimedia content has made accurately retrieving relevant videos from large collections increasingly challenging. Recent advancements in text-video retrieval have focused on cross-modal interactions, large-scale foundation model training, and probabilistic modeling, yet often neglect the crucial user perspective, leading to discrepancies between user queries and the content retrieved. To address this, we introduce MERLIN (Multimodal Embedding Refinement via LLM-based Iterative Navigation), a novel, training-free pipeline that leverages Large Language Models (LLMs) for iterative feedback learning. MERLIN refines query embeddings from a user perspective, enhancing alignment between queries and video content through a dynamic question answering process. Experimental results on datasets like MSR-VTT, MSVD, and ActivityNet demonstrate that MERLIN substantially improves Recall@1, outperforming existing systems and confirming the benefits of integrating LLMs into multimodal retrieval systems for more responsive and context-aware multimedia retrieval.

cs.CL↗