Sanghyeok Lee

Jinyoung Park, Sanghyeok Lee, Sihyeon Kim et al.

Transformers have shown superior performance on various computer vision tasks with their capabilities to capture long-range dependencies. Despite the success, it is challenging to directly apply Transformers on point clouds due to their quadratic cost in the number of points. In this paper, we present a Self-Positioning point-based Transformer (SPoTr), which is designed to capture both local and global shape contexts with reduced complexity. Specifically, this architecture consists of local self-attention and self-positioning point-based global cross-attention. The self-positioning points, adaptively located based on the input shape, consider both spatial and semantic information with disentangled attention to improve expressive power. With the self-positioning points, we propose a novel global cross-attention mechanism for point clouds, which improves the scalability of global self-attention by allowing the attention module to compute attention weights with only a small set of self-positioning points. Experiments show the effectiveness of SPoTr on three point cloud tasks such as shape classification, part segmentation, and scene segmentation. In particular, our proposed model achieves an accuracy gain of 2.6% over the previous best models on shape classification with ScanObjectNN. We also provide qualitative analyses to demonstrate the interpretability of self-positioning points. The code of SPoTr is available at https://github.com/mlvlab/SPoTr.

Dongjun Lee, Seokwon Song, Jihee Suh et al.

In recent years, prompt tuning has proven effective in adapting pre-trained vision-language models to downstream tasks. These methods aim to adapt the pre-trained models by introducing learnable prompts while keeping pre-trained weights frozen. However, learnable prompts can affect the internal representation within the self-attention module, which may negatively impact performance variance and generalization, especially in data-deficient settings. To address these issues, we propose a novel approach, Read-only Prompt Optimization (RPO). RPO leverages masked attention to prevent the internal representation shift in the pre-trained model. Further, to facilitate the optimization of RPO, the read-only prompts are initialized based on special tokens of the pre-trained model. Our extensive experiments demonstrate that RPO outperforms CLIP and CoCoOp in base-to-new generalization and domain generalization while displaying better robustness. Also, the proposed method achieves better generalization on extremely data-deficient settings, while improving parameter efficiency and computational overhead. Code is available at https://github.com/mlvlab/RPO.

Sanghyeok Lee, Minkyu Jeon, Injae Kim et al.

Data augmentation is key to improving the generalization ability of deep learning models. Mixup is a simple and widely-used data augmentation technique that has proven effective in alleviating the problems of overfitting and data scarcity. Also, recent studies of saliency-aware Mixup in the image domain show that preserving discriminative parts is beneficial to improving the generalization performance. However, these Mixup-based data augmentations are underexplored in 3D vision, especially in point clouds. In this paper, we propose SageMix, a saliency-guided Mixup for point clouds to preserve salient local structures. Specifically, we extract salient regions from two point clouds and smoothly combine them into one continuous shape. With a simple sequential sampling by re-weighted saliency scores, SageMix preserves the local structure of salient regions. Extensive experiments demonstrate that the proposed method consistently outperforms existing Mixup methods in various benchmark point cloud datasets. With PointNet++, our method achieves an accuracy gain of 2.6% and 4.0% over standard training in 3D Warehouse dataset (MN40) and ScanObjectNN, respectively. In addition to generalization performance, SageMix improves robustness and uncertainty calibration. Moreover, when adopting our method to various tasks including part segmentation and standard 2D image classification, our method achieves competitive performance.

Juhan Cha, Minseok Joo, Jihwan Park et al.

Recent advancements in 3D object detection have benefited from multi-modal information from the multi-view cameras and LiDAR sensors. However, the inherent disparities between the modalities pose substantial challenges. We observe that existing multi-modal 3D object detection methods heavily rely on the LiDAR sensor, treating the camera as an auxiliary modality for augmenting semantic details. This often leads to not only underutilization of camera data but also significant performance degradation in scenarios where LiDAR data is unavailable. Additionally, existing fusion methods overlook the detrimental impact of sensor noise induced by environmental changes, on detection performance. In this paper, we propose MEFormer to address the LiDAR over-reliance problem by harnessing critical information for 3D object detection from every available modality while concurrently safeguarding against corrupted signals during the fusion process. Specifically, we introduce Modality Agnostic Decoding (MOAD) that extracts geometric and semantic features with a shared transformer decoder regardless of input modalities and provides promising improvement with a single modality as well as multi-modality. Additionally, our Proximity-based Modality Ensemble (PME) module adaptively utilizes the strengths of each modality depending on the environment while mitigating the effects of a noisy sensor. Our MEFormer achieves state-of-the-art performance of 73.9% NDS and 71.5% mAP in the nuScenes validation set. Extensive analyses validate that our MEFormer improves robustness against challenging conditions such as sensor malfunctions or environmental changes. The source code is available at https://github.com/hanchaa/MEFormer

Joonmyung Choi, Sanghyeok Lee, Jaewon Chu et al.

Video Transformers have become the prevalent solution for various video downstream tasks with superior expressive power and flexibility. However, these video transformers suffer from heavy computational costs induced by the massive number of tokens across the entire video frames, which has been the major barrier to training the model. Further, the patches irrelevant to the main contents, e.g., backgrounds, degrade the generalization performance of models. To tackle these issues, we propose training free token merging for lightweight video Transformer (vid-TLDR) that aims to enhance the efficiency of video Transformers by merging the background tokens without additional training. For vid-TLDR, we introduce a novel approach to capture the salient regions in videos only with the attention map. Further, we introduce the saliency-aware token merging strategy by dropping the background tokens and sharpening the object scores. Our experiments show that vid-TLDR significantly mitigates the computational complexity of video Transformers while achieving competitive performance compared to the base model without vid-TLDR. Code is available at https://github.com/mlvlab/vid-TLDR.

Sanghyeok Lee, Joonmyung Choi, Hyunwoo J. Kim

Vision Transformer (ViT) has emerged as a prominent backbone for computer vision. For more efficient ViTs, recent works lessen the quadratic cost of the self-attention layer by pruning or fusing the redundant tokens. However, these works faced the speed-accuracy trade-off caused by the loss of information. Here, we argue that token fusion needs to consider diverse relations between tokens to minimize information loss. In this paper, we propose a Multi-criteria Token Fusion (MCTF), that gradually fuses the tokens based on multi-criteria (e.g., similarity, informativeness, and size of fused tokens). Further, we utilize the one-step-ahead attention, which is the improved approach to capture the informativeness of the tokens. By training the model equipped with MCTF using a token reduction consistency, we achieve the best speed-accuracy trade-off in the image classification (ImageNet1K). Experimental results prove that MCTF consistently surpasses the previous reduction methods with and without training. Specifically, DeiT-T and DeiT-S with MCTF reduce FLOPs by about 44% while improving the performance (+0.5%, and +0.3%) over the base model, respectively. We also demonstrate the applicability of MCTF in various Vision Transformers (e.g., T2T-ViT, LV-ViT), achieving at least 31% speedup without performance degradation. Code is available at https://github.com/mlvlab/MCTF.

Dohwan Ko, Jinyoung Park, Seoung Choi et al.

Mixture-of-Experts (MoE) has emerged as an effective approach to reduce the computational overhead of Transformer architectures by sparsely activating a subset of parameters for each token while preserving high model capacity. This paradigm has recently been extended to Vision-Language Models (VLMs), enabling scalable multi-modal understanding with reduced computational cost. However, the widely adopted deterministic top-K routing mechanism may overlook more optimal expert combinations and lead to expert overfitting. To address this limitation and improve the diversity of expert selection, we propose MoE-GRPO, a reinforcement learning (RL)-based framework for optimizing expert routing in MoE-based VLMs. Specifically, we formulate expert selection as a sequential decision-making problem and optimize it using Group Relative Policy Optimization (GRPO), allowing the model to learn adaptive expert routing policies through exploration and reward-based feedback. Furthermore, we introduce a modality-aware router guidance that enhances training stability and efficiency by discouraging the router from exploring experts that are infrequently activated for a given modality. Extensive experiments on multi-modal image and video benchmarks show that MoE-GRPO consistently outperforms standard top-K routing and its variants by promoting more diverse expert selection, thereby mitigating expert overfitting and enabling a task-level expert specialization.

Joonmyung Choi, Sanghyeok Lee, Jongha Kim et al.

Recent advances in vision-language models have demonstrated remarkable performance across diverse multi-modal tasks, including document question answering that leverages structured visual cues from text, tables, and figures. However, unlike natural images, document images contain large backgrounds and only sparse supporting evidence, leading to the inefficient consumption of substantial computational resources, especially for long documents. We observe that existing token-reduction methods for natural images and videos fall short in utilizing the structural sparsity unique to documents. To address this, we propose DocPrune, a training-free and progressive document token pruning framework designed for efficient long-document understanding. The proposed method preserves only the essential tokens for the task while removing unnecessary ones, such as background or question-irrelevant tokens. Moreover, it automatically selects the appropriate layers to initiate token pruning based on the model's level of comprehension. Our experiments on the M3DocRAG show that DocPrune improves throughput by 3.0x and 3.3x in the encoder and decoder, respectively, while boosting the F1 score by +1.0, achieving both higher accuracy and efficiency without any additional training.

Sanghyeok Lee, Joonmyung Choi, Hyunwoo J. Kim

For the deployment of neural networks in resource-constrained environments, prior works have built lightweight architectures with convolution and attention for capturing local and global dependencies, respectively. Recently, the state space model (SSM) has emerged as an effective operation for global interaction with its favorable linear computational cost in the number of tokens. To harness the efficacy of SSM, we introduce Efficient Vision Mamba (EfficientViM), a novel architecture built on hidden state mixer-based state space duality (HSM-SSD) that efficiently captures global dependencies with further reduced computational cost. With the observation that the runtime of the SSD layer is driven by the linear projections on the input sequences, we redesign the original SSD layer to perform the channel mixing operation within compressed hidden states in the HSM-SSD layer. Additionally, we propose multi-stage hidden state fusion to reinforce the representation power of hidden states and provide the design to alleviate the bottleneck caused by the memory-bound operations. As a result, the EfficientViM family achieves a new state-of-the-art speed-accuracy trade-off on ImageNet-1k, offering up to a 0.7% performance improvement over the second-best model SHViT with faster speed. Further, we observe significant improvements in throughput and accuracy compared to prior works, when scaling images or employing distillation training. Code is available at https://github.com/mlvlab/EfficientViM.

Jongha Kim, Minseong Bae, Sanghyeok Lee et al.

Table images present unique challenges for effective and efficient understanding due to the need for question-specific focus and the presence of redundant background regions. Existing Multimodal Large Language Model (MLLM) approaches often overlook these characteristics, resulting in uninformative and redundant visual representations. To address these issues, we aim to generate visual features that are both informative and compact to improve table understanding. We first propose progressive question conditioning, which injects the question into Vision Transformer layers with gradually increasing frequency, considering each layer's capacity to handle additional information, to generate question-aware visual features. To reduce redundancy, we introduce a pruning strategy that discards background tokens, thereby improving efficiency. To mitigate information loss from pruning, we further propose token focusing, a training strategy that encourages the model to concentrate essential information in the retained tokens. By combining these approaches, we present TabFlash, an efficient and effective MLLM for table understanding. TabFlash achieves state-of-the-art performance, outperforming both open-source and proprietary MLLMs, while requiring 27% less FLOPs and 30% less memory usage compared to the second-best MLLM.

Minsuk Ji, Sanghyeok Lee, Namhyuk Ahn

Despite their impressive realism, modern text-to-image models still struggle with compositionality, often failing to render accurate object counts, attributes, and spatial relations. To address this challenge, we present a training-free framework that combines an object-centric approach with self-refinement to improve layout faithfulness while preserving aesthetic quality. Specifically, we leverage large language models (LLMs) to synthesize explicit layouts from input prompts, and we inject these layouts into the image generation process, where a object-centric vision-language model (VLM) judge reranks multiple candidates to select the most prompt-aligned outcome iteratively. By unifying explicit layout-grounding with self-refine-based inference-time scaling, our framework achieves stronger scene alignment with prompts compared to recent text-to-image models. The code are available at https://github.com/gcl-inha/ReFocus.

Joonmyung Choi, Sanghyeok Lee, Byungoh Ko et al.

Transformers have demonstrated remarkable success across vision, language, and video. Yet, increasing task complexity has led to larger models and more tokens, raising the quadratic cost of self-attention and the overhead of GPU memory access. To reduce the computation cost of self-attention, prior work has proposed token compression techniques that drop redundant or less informative tokens. Meanwhile, fused attention kernels such as FlashAttention have been developed to alleviate memory overhead by avoiding attention map construction and its associated I/O to HBM. This, however, makes it incompatible with most training-free token compression methods, which rely on attention maps to determine token importance. Here, we propose Representation Shift, a training-free, model-agnostic metric that measures the degree of change in each token's representation. This seamlessly integrates token compression with FlashAttention, without attention maps or retraining. Our method further generalizes beyond Transformers to CNNs and state space models. Extensive experiments show that Representation Shift enables effective token compression compatible with FlashAttention, yielding significant speedups of up to 5.5% and 4.4% in video-text retrieval and video QA, respectively. Code is available at https://github.com/mlvlab/Representation-Shift.

Jinyoung Park, Sanghyeok Lee, Omar Zia Khan et al.

Graph Retrieval-Augmented Generation (GraphRAG) has been successfully applied in various knowledge-intensive question answering tasks by organizing external knowledge into structured graphs of entities and relations. It enables large language models (LLMs) to perform complex reasoning beyond text-chunk retrieval. Recent works have employed reinforcement learning (RL) to train agentic GraphRAG frameworks that perform iterative interactions between LLMs and knowledge graphs. However, existing RL-based frameworks such as Graph-R1 suffer from two key limitations: (1) they primarily depend on semantic similarity for retrieval, often overlooking the underlying graph structure, and (2) they rely on sparse, outcome-level rewards, failing to capture the quality of intermediate retrieval steps and their dependencies. To address these limitations, we propose ProGraph-R1, a progress-aware agentic framework for graph-based retrieval and multi-step reasoning. ProGraph-R1 introduces a structure-aware hypergraph retrieval mechanism that jointly considers semantic relevance and graph connectivity, encouraging coherent traversal along multi-hop reasoning paths. We also design a progress-based step-wise policy optimization, which provides dense learning signals by modulating advantages according to intermediate reasoning progress within a graph, rather than relying solely on final outcomes. Experiments on multi-hop question answering benchmarks demonstrate that ProGraph-R1 consistently improves reasoning accuracy and generation quality over existing GraphRAG methods.

Kyungryul Back, Seongbeom Park, Milim Kim et al.

Large Vision-Language Models (LVLMs) have recently shown promising results on various multimodal tasks, even achieving human-comparable performance in certain cases. Nevertheless, LVLMs remain prone to hallucinations -- they often rely heavily on a single modality or memorize training data without properly grounding their outputs. To address this, we propose a training-free, tri-layer contrastive decoding with watermarking, which proceeds in three steps: (1) select a mature layer and an amateur layer among the decoding layers, (2) identify a pivot layer using a watermark-related question to assess whether the layer is visually well-grounded, and (3) apply tri-layer contrastive decoding to generate the final output. Experiments on public benchmarks such as POPE, MME and AMBER demonstrate that our method achieves state-of-the-art performance in reducing hallucinations in LVLMs and generates more visually grounded responses.

Jihwan Park, Taehoon song, Sanghyeok Lee et al.

Vision-Language Models (VLMs) have been widely used in various visual recognition tasks due to their remarkable generalization capabilities. As these models grow in size and complexity, fine-tuning becomes costly, emphasizing the need to reuse adaptation knowledge from 'weaker' models to efficiently enhance 'stronger' ones. However, existing adaptation transfer methods exhibit limited transferability across models due to their model-specific design and high computational demands. To tackle this, we propose Transferable Model-agnostic adapter (TransMiter), a light-weight adapter that improves vision-language models 'without backpropagation'. TransMiter captures the knowledge gap between pre-trained and fine-tuned VLMs, in an 'unsupervised' manner. Once trained, this knowledge can be seamlessly transferred across different models without the need for backpropagation. Moreover, TransMiter consists of only a few layers, inducing a negligible additional inference cost. Notably, supplementing the process with a few labeled data further yields additional performance gain, often surpassing a fine-tuned stronger model, with a marginal training cost. Experimental results and analyses demonstrate that TransMiter effectively and efficiently transfers adaptation knowledge while preserving generalization abilities across VLMs of different sizes and architectures in visual recognition tasks.

Sihyeon Kim, Sanghyeok Lee, Dasol Hwang et al.

Despite the extensive usage of point clouds in 3D vision, relatively limited data are available for training deep neural networks. Although data augmentation is a standard approach to compensate for the scarcity of data, it has been less explored in the point cloud literature. In this paper, we propose a simple and effective augmentation method called PointWOLF for point cloud augmentation. The proposed method produces smoothly varying non-rigid deformations by locally weighted transformations centered at multiple anchor points. The smooth deformations allow diverse and realistic augmentations. Furthermore, in order to minimize the manual efforts to search the optimal hyperparameters for augmentation, we present AugTune, which generates augmented samples of desired difficulties producing targeted confidence scores. Our experiments show our framework consistently improves the performance for both shape classification and part segmentation tasks. Particularly, with PointNet++, PointWOLF achieves the state-of-the-art 89.7 accuracy on shape classification with the real-world ScanObjectNN dataset.