Dongyoon Wee

Jeongseok Hyun, Myunggu Kang, Dongyoon Wee et al.

In existing joint detection and tracking methods, pairwise relational features are used to match previous tracklets to current detections. However, the features may not be discriminative enough for a tracker to identify a target from a large number of detections. Selecting only high-scored detections for tracking may lead to missed detections whose confidence score is low. Consequently, in the online setting, this results in disconnections of tracklets which cannot be recovered. In this regard, we present Sparse Graph Tracker (SGT), a novel online graph tracker using higher-order relational features which are more discriminative by aggregating the features of neighboring detections and their relations. SGT converts video data into a graph where detections, their connections, and the relational features of two connected nodes are represented by nodes, edges, and edge features, respectively. The strong edge features allow SGT to track targets with tracking candidates selected by top-K scored detections with large K. As a result, even low-scored detections can be tracked, and the missed detections are also recovered. The robustness of K value is shown through the extensive experiments. In the MOT16/17/20 and HiEve Challenge, SGT outperforms the state-of-the-art trackers with real-time inference speed. Especially, a large improvement in MOTA is shown in the MOT20 and HiEve Challenge. Code is available at https://github.com/HYUNJS/SGT.

Taeoh Kim, Jinhyung Kim, Minho Shim et al.

Data augmentation has recently emerged as an essential component of modern training recipes for visual recognition tasks. However, data augmentation for video recognition has been rarely explored despite its effectiveness. Few existing augmentation recipes for video recognition naively extend the image augmentation methods by applying the same operations to the whole video frames. Our main idea is that the magnitude of augmentation operations for each frame needs to be changed over time to capture the real-world video's temporal variations. These variations should be generated as diverse as possible using fewer additional hyper-parameters during training. Through this motivation, we propose a simple yet effective video data augmentation framework, DynaAugment. The magnitude of augmentation operations on each frame is changed by an effective mechanism, Fourier Sampling that parameterizes diverse, smooth, and realistic temporal variations. DynaAugment also includes an extended search space suitable for video for automatic data augmentation methods. DynaAugment experimentally demonstrates that there are additional performance rooms to be improved from static augmentations on diverse video models. Specifically, we show the effectiveness of DynaAugment on various video datasets and tasks: large-scale video recognition (Kinetics-400 and Something-Something-v2), small-scale video recognition (UCF- 101 and HMDB-51), fine-grained video recognition (Diving-48 and FineGym), video action segmentation on Breakfast, video action localization on THUMOS'14, and video object detection on MOT17Det. DynaAugment also enables video models to learn more generalized representation to improve the model robustness on the corrupted videos.

Pilhyeon Lee, Taeoh Kim, Minho Shim et al.

Temporal action detection aims to predict the time intervals and the classes of action instances in the video. Despite the promising performance, existing two-stream models exhibit slow inference speed due to their reliance on computationally expensive optical flow. In this paper, we introduce a decomposed cross-modal distillation framework to build a strong RGB-based detector by transferring knowledge of the motion modality. Specifically, instead of direct distillation, we propose to separately learn RGB and motion representations, which are in turn combined to perform action localization. The dual-branch design and the asymmetric training objectives enable effective motion knowledge transfer while preserving RGB information intact. In addition, we introduce a local attentive fusion to better exploit the multimodal complementarity. It is designed to preserve the local discriminability of the features that is important for action localization. Extensive experiments on the benchmarks verify the effectiveness of the proposed method in enhancing RGB-based action detectors. Notably, our framework is agnostic to backbones and detection heads, bringing consistent gains across different model combinations.

Jaehyeok Kim, Dongyoon Wee, Dan Xu

We introduce You Only Train Once (YOTO), a dynamic human generation framework, which performs free-viewpoint rendering of different human identities with distinct motions, via only one-time training from monocular videos. Most prior works for the task require individualized optimization for each input video that contains a distinct human identity, leading to a significant amount of time and resources for the deployment, thereby impeding the scalability and the overall application potential of the system. In this paper, we tackle this problem by proposing a set of learnable identity codes to expand the capability of the framework for multi-identity free-viewpoint rendering, and an effective pose-conditioned code query mechanism to finely model the pose-dependent non-rigid motions. YOTO optimizes neural radiance fields (NeRF) by utilizing designed identity codes to condition the model for learning various canonical T-pose appearances in a single shared volumetric representation. Besides, our joint learning of multiple identities within a unified model incidentally enables flexible motion transfer in high-quality photo-realistic renderings for all learned appearances. This capability expands its potential use in important applications, including Virtual Reality. We present extensive experimental results on ZJU-MoCap and PeopleSnapshot to clearly demonstrate the effectiveness of our proposed model. YOTO shows state-of-the-art performance on all evaluation metrics while showing significant benefits in training and inference efficiency as well as rendering quality. The code and model will be made publicly available soon.

Jinhyung Kim, Taeoh Kim, Minho Shim et al.

Recent self-supervised video representation learning methods focus on maximizing the similarity between multiple augmented views from the same video and largely rely on the quality of generated views. However, most existing methods lack a mechanism to prevent representation learning from bias towards static information in the video. In this paper, we propose frequency augmentation (FreqAug), a spatio-temporal data augmentation method in the frequency domain for video representation learning. FreqAug stochastically removes specific frequency components from the video so that learned representation captures essential features more from the remaining information for various downstream tasks. Specifically, FreqAug pushes the model to focus more on dynamic features rather than static features in the video via dropping spatial or temporal low-frequency components. To verify the generality of the proposed method, we experiment with FreqAug on multiple self-supervised learning frameworks along with standard augmentations. Transferring the improved representation to five video action recognition and two temporal action localization downstream tasks shows consistent improvements over baselines.

Nicolas Monet, Dongyoon Wee

This technical report introduces our solution, MEEV, proposed to the EgoBody Challenge at ECCV 2022. Captured from head-mounted devices, the dataset consists of human body shape and motion of interacting people. The EgoBody dataset has challenges such as occluded body or blurry image. In order to overcome the challenges, MEEV is designed to exploit multiscale features for rich spatial information. Besides, to overcome the limited size of dataset, the model is pre-trained with the dataset aggregated 2D and 3D pose estimation datasets. Achieving 82.30 for MPJPE and 92.93 for MPVPE, MEEV has won the EgoBody Challenge at ECCV 2022, which shows the effectiveness of the proposed method. The code is available at https://github.com/clovaai/meev

Jinsung Lee, Taeoh Kim, Inwoong Lee et al.

Video action detection (VAD) aims to detect actors and classify their actions in a video. We figure that VAD suffers more from classification rather than localization of actors. Hence, we analyze how prevailing methods form features for classification and find that they prioritize actor regions, yet often overlooking the essential contextual information necessary for accurate classification. Accordingly, we propose to reduce the bias toward actor and encourage paying attention to the context that is relevant to each action class. By assigning a class-dedicated query to each action class, our model can dynamically determine where to focus for effective classification. The proposed model demonstrates superior performance on three challenging benchmarks with significantly fewer parameters and less computation.

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.

Geonho Cha, Chaehun Shin, Sungroh Yoon et al.

To estimate the volume density and color of a 3D point in the multi-view image-based rendering, a common approach is to inspect the consensus existence among the given source image features, which is one of the informative cues for the estimation procedure. To this end, most of the previous methods utilize equally-weighted aggregation features. However, this could make it hard to check the consensus existence when some outliers, which frequently occur by occlusions, are included in the source image feature set. In this paper, we propose a novel source-view-wise feature aggregation method, which facilitates us to find out the consensus in a robust way by leveraging local structures in the feature set. We first calculate the source-view-wise distance distribution for each source feature for the proposed aggregation. After that, the distance distribution is converted to several similarity distributions with the proposed learnable similarity mapping functions. Finally, for each element in the feature set, the aggregation features are extracted by calculating the weighted means and variances, where the weights are derived from the similarity distributions. In experiments, we validate the proposed method on various benchmark datasets, including synthetic and real image scenes. The experimental results demonstrate that incorporating the proposed features improves the performance by a large margin, resulting in the state-of-the-art performance.

Minho Shim, Taeoh Kim, Jinhyung Kim et al.

Summarizing a video requires a diverse understanding of the video, ranging from recognizing scenes to evaluating how much each frame is essential enough to be selected as a summary. Self-supervised learning (SSL) is acknowledged for its robustness and flexibility to multiple downstream tasks, but the video SSL has not shown its value for dense understanding tasks like video summarization. We claim an unsupervised autoencoder with sufficient self-supervised learning does not need any extra downstream architecture design or fine-tuning weights to be utilized as a video summarization model. The proposed method to evaluate the importance score of each frame takes advantage of the reconstruction score of the autoencoder's decoder. We evaluate the method in major unsupervised video summarization benchmarks to show its effectiveness under various experimental settings.

Jaehyeok Kim, Dongyoon Wee, Dan Xu

This paper introduces Motion-oriented Compositional Neural Radiance Fields (MoCo-NeRF), a framework designed to perform free-viewpoint rendering of monocular human videos via novel non-rigid motion modeling approach. In the context of dynamic clothed humans, complex cloth dynamics generate non-rigid motions that are intrinsically distinct from skeletal articulations and critically important for the rendering quality. The conventional approach models non-rigid motions as spatial (3D) deviations in addition to skeletal transformations. However, it is either time-consuming or challenging to achieve optimal quality due to its high learning complexity without a direct supervision. To target this problem, we propose a novel approach of modeling non-rigid motions as radiance residual fields to benefit from more direct color supervision in the rendering and utilize the rigid radiance fields as a prior to reduce the complexity of the learning process. Our approach utilizes a single multiresolution hash encoding (MHE) to concurrently learn the canonical T-pose representation from rigid skeletal motions and the radiance residual field for non-rigid motions. Additionally, to further improve both training efficiency and usability, we extend MoCo-NeRF to support simultaneous training of multiple subjects within a single framework, thanks to our effective design for modeling non-rigid motions. This scalability is achieved through the integration of a global MHE and learnable identity codes in addition to multiple local MHEs. We present extensive results on ZJU-MoCap and MonoCap, clearly demonstrating state-of-the-art performance in both single- and multi-subject settings. The code and model will be made publicly available at the project page: https://stevejaehyeok.github.io/publications/moco-nerf.

Seonghoon Yu, Ilchae Jung, Byeongju Han et al.

Referring image segmentation (RIS) requires dense vision-language interactions between visual pixels and textual words to segment objects based on a given description. However, commonly adapted dual-encoders in RIS, e.g., Swin transformer and BERT (uni-modal encoders) or CLIP (a multi-modal dual-encoder), lack dense multi-modal interactions during pre-training, leading to a gap with a pixel-level RIS task. To bridge this gap, existing RIS methods often rely on multi-modal fusion modules that interact two encoders, but this approach leads to high computational costs. In this paper, we present a novel RIS method with a single-encoder, i.e., BEiT-3, maximizing the potential of shared self-attention across all framework components. This enables seamless interactions of two modalities from input to final prediction, producing granularly aligned multi-modal features. Furthermore, we propose lightweight yet effective decoder modules, a Shared FPN and a Shared Mask Decoder, which contribute to the high efficiency of our model. Our simple baseline with a single encoder achieves outstanding performances on the RIS benchmark datasets while maintaining computational efficiency, compared to the most recent SoTA methods based on dual-encoders.

Geonho Cha, Ho-Deok Jang, Dongyoon Wee

Managing the dynamic regions in the photometric loss formulation has been a main issue for handling the self-supervised depth estimation problem. Most previous methods have alleviated this issue by removing the dynamic regions in the photometric loss formulation based on the masks estimated from another module, making it difficult to fully utilize the training images. In this paper, to handle this problem, we propose an isometric self-sample-based learning (ISSL) method to fully utilize the training images in a simple yet effective way. The proposed method provides additional supervision during training using self-generated images that comply with pure static scene assumption. Specifically, the isometric self-sample generator synthesizes self-samples for each training image by applying random rigid transformations on the estimated depth. Thus both the generated self-samples and the corresponding training image always follow the static scene assumption. We show that plugging our ISSL module into several existing models consistently improves the performance by a large margin. In addition, it also boosts the depth accuracy over different types of scene, i.e., outdoor scenes (KITTI and Make3D) and indoor scene (NYUv2), validating its high effectiveness.

Jiwoo Chung, Sangeek Hyun, MinKyu Lee et al.

Diffusion models are a strong backbone for visual generation, but their inherently sequential denoising process leads to slow inference. Previous methods accelerate sampling by caching and reusing intermediate outputs based on feature distances between adjacent timesteps. However, existing caching strategies typically rely on raw feature differences that entangle content and noise. This design overlooks spectral evolution, where low-frequency structure appears early and high-frequency detail is refined later. We introduce Spectral-Evolution-Aware Cache (SeaCache), a training-free cache schedule that bases reuse decisions on a spectrally aligned representation. Through theoretical and empirical analysis, we derive a Spectral-Evolution-Aware (SEA) filter that preserves content-relevant components while suppressing noise. Employing SEA-filtered input features to estimate redundancy leads to dynamic schedules that adapt to content while respecting the spectral priors underlying the diffusion model. Extensive experiments on diverse visual generative models and the baselines show that SeaCache achieves state-of-the-art latency-quality trade-offs.

Geo Ahn, Inwoong Lee, Taeoh Kim et al.

We study Compositional Video Understanding (CVU), where models must recognize verbs and objects and compose them to generalize to unseen combinations. We find that existing Zero-Shot Compositional Action Recognition (ZS-CAR) models fail primarily due to an overlooked failure mode: object-driven verb shortcuts. Through systematic analysis, we show that this behavior arises from two intertwined factors: severe sparsity and skewness of compositional supervision, and the asymmetric learning difficulty between verbs and objects. As training progresses, the existing ZS-CAR model increasingly ignores visual evidence and overfits to co-occurrence statistics. Consequently, the existing model does not gain the benefit of compositional recognition in unseen verb-object compositions. To address this, we propose RCORE, a simple and effective framework that enforces temporally grounded verb learning. RCORE introduces (i) a composition-aware augmentation that diversifies verb-object combinations without corrupting motion cues, and (ii) a temporal order regularization loss that penalizes shortcut behaviors by explicitly modeling temporal structure. Across two benchmarks, Sth-com and our newly constructed EK100-com, RCORE significantly improves unseen composition accuracy, reduces reliance on co-occurrence bias, and achieves consistently positive compositional gaps. Our findings reveal object-driven shortcuts as a critical limiting factor in ZS-CAR and demonstrate that addressing them is essential for robust compositional video understanding.

Geuntaek Lim, Minho Shim, Sungjune Park et al.

The inherent complexity of video understanding makes it difficult to attribute whether performance gains stem from visual perception, linguistic reasoning, or knowledge priors. While many benchmarks have emerged to assess high-level reasoning, the essential criteria that constitute video understanding remain largely overlooked. Instead of introducing yet another benchmark, we take a step back to re-examine the current landscape of video understanding. In this work, we provide Video-Oasis, a sustainable diagnostic suite designed to systematically evaluate existing evaluations and distill spatio-temporal challenges for video understanding. Our analysis reveals two critical findings: (1) 54% of existing benchmark samples are solvable without visual input or temporal context, and (2) on the remaining samples, state-of-the-art models exhibit performance barely exceeding random guessing. To bridge this gap, we investigate which algorithmic design choices contribute to robust video understanding, providing practical guidelines for future research. We hope our work serves as a standard guideline for benchmark construction and the rigorous evaluation of architecture development. Code is available at https://github.com/sejong-rcv/Video-Oasis.

Su Ho Han, Jeongseok Hyun, Pilhyeon Lee et al.

Multimodal large language models (MLLMs) demonstrate strong video understanding by attending to visual tokens relevant to textual queries. To directly adapt this for localization in a training-free manner, we cast video reasoning segmentation as a video QA task and extract attention maps via rollout mechanism. However, raw attention maps are noisy and poorly aligned with object regions. We propose Decomposed Attention Fusion (DecAF), which refines these maps through two mechanisms: (1) contrastive object-background fusion and (2) complementary video-frame fusion. This method suppresses irrelevant activations and enhances object-focused cues, enabling direct conversion of attention maps into coarse segmentation masks. In addition, we introduce attention-guided SAM2 prompting for obtaining fine-grained masks. Unlike existing methods that jointly train MLLMs with SAM, our method operates entirely without retraining. DecAF outperforms training-free methods and achieves performance comparable to training-based methods on both referring and reasoning VOS benchmarks. The code will be available at https://github.com/HYUNJS/DecAF.

Jeongseok Hyun, Sukjun Hwang, Su Ho Han et al.

Video large language models (LLMs) achieve strong video understanding by leveraging a large number of spatio-temporal tokens, but suffer from quadratic computational scaling with token count. To address this, we propose a training-free spatio-temporal token merging method, named STTM. Our key insight is to exploit local spatial and temporal redundancy in video data which has been overlooked in prior work. STTM first transforms each frame into multi-granular spatial tokens using a coarse-to-fine search over a quadtree structure, then performs directed pairwise merging across the temporal dimension. This decomposed merging approach outperforms existing token reduction methods across six video QA benchmarks. Notably, STTM achieves a 2$\times$ speed-up with only a 0.5% accuracy drop under a 50% token budget, and a 3$\times$ speed-up with just a 2% drop under a 30% budget. Moreover, STTM is query-agnostic, allowing KV cache reuse across different questions for the same video. The project page is available at https://www.jshyun.me/projects/sttm.

WonJun Moon, Cheol-Ho Cho, Woojin Jun et al.

In a retrieval system, simultaneously achieving search accuracy and efficiency is inherently challenging. This challenge is particularly pronounced in partially relevant video retrieval (PRVR), where incorporating more diverse context representations at varying temporal scales for each video enhances accuracy but increases computational and memory costs. To address this dichotomy, we propose a prototypical PRVR framework that encodes diverse contexts within a video into a fixed number of prototypes. We then introduce several strategies to enhance text association and video understanding within the prototypes, along with an orthogonal objective to ensure that the prototypes capture a diverse range of content. To keep the prototypes searchable via text queries while accurately encoding video contexts, we implement cross- and uni-modal reconstruction tasks. The cross-modal reconstruction task aligns the prototypes with textual features within a shared space, while the uni-modal reconstruction task preserves all video contexts during encoding. Additionally, we employ a video mixing technique to provide weak guidance to further align prototypes and associated textual representations. Extensive evaluations on TVR, ActivityNet-Captions, and QVHighlights validate the effectiveness of our approach without sacrificing efficiency.

Changwoon Choi, Jeongjun Kim, Geonho Cha et al.

Recent works on dynamic 3D neural field reconstruction assume the input from synchronized multi-view videos whose poses are known. The input constraints are often not satisfied in real-world setups, making the approach impractical. We show that unsynchronized videos from unknown poses can generate dynamic neural fields as long as the videos capture human motion. Humans are one of the most common dynamic subjects captured in videos, and their shapes and poses can be estimated using state-of-the-art libraries. While noisy, the estimated human shape and pose parameters provide a decent initialization point to start the highly non-convex and under-constrained problem of training a consistent dynamic neural representation. Given the shape and pose parameters of humans in individual frames, we formulate methods to calculate the time offsets between videos, followed by camera pose estimations that analyze the 3D joint positions. Then, we train the dynamic neural fields employing multiresolution grids while we concurrently refine both time offsets and camera poses. The setup still involves optimizing many parameters; therefore, we introduce a robust progressive learning strategy to stabilize the process. Experiments show that our approach achieves accurate spatio-temporal calibration and high-quality scene reconstruction in challenging conditions.

Taewoo Kim, Suwoong Yeom, Jaehyun Pyun et al.

Reconstructing dynamic scenes with complex human-object interactions is a fundamental challenge in computer vision and graphics. Existing Gaussian Splatting methods either rely on human pose priors while neglecting dynamic objects, or approximate all motions within a single field, limiting their ability to capture interaction-rich dynamics. To address this gap, we propose Human-Object Interaction Gaussian Splatting (HOIGS), which explicitly models interaction-induced deformation between humans and objects through a cross-attention-based HOI module. Distinct deformation baselines are employed to extract features: HexPlane for humans and Cubic Hermite Spline (CHS) for objects. By integrating these heterogeneous features, HOIGS effectively captures interdependent motions and improves deformation estimation in scenarios involving occlusion, contact, and object manipulation. Comprehensive experiments on multiple datasets demonstrate that our method consistently outperforms state-of-the-art human-centric and 4D Gaussian approaches, highlighting the importance of explicitly modeling human-object interactions for high-fidelity reconstruction.

Jungho Lee, Suhwan Cho, Taeoh Kim et al.

3D Gaussian Splatting (3DGS) has attracted significant attention for its high-quality novel view rendering, inspiring research to address real-world challenges. While conventional methods depend on sharp images for accurate scene reconstruction, real-world scenarios are often affected by defocus blur due to finite depth of field, making it essential to account for realistic 3D scene representation. In this study, we propose CoCoGaussian, a Circle of Confusion-aware Gaussian Splatting that enables precise 3D scene representation using only defocused images. CoCoGaussian addresses the challenge of defocus blur by modeling the Circle of Confusion (CoC) through a physically grounded approach based on the principles of photographic defocus. Exploiting 3D Gaussians, we compute the CoC diameter from depth and learnable aperture information, generating multiple Gaussians to precisely capture the CoC shape. Furthermore, we introduce a learnable scaling factor to enhance robustness and provide more flexibility in handling unreliable depth in scenes with reflective or refractive surfaces. Experiments on both synthetic and real-world datasets demonstrate that CoCoGaussian achieves state-of-the-art performance across multiple benchmarks.

Sangmin Kim, Minhyuk Hwang, Geonho Cha et al.

Recent advances in 3D foundation models have led to growing interest in reconstructing humans and their surrounding environments. However, most existing approaches focus on monocular inputs, and extending them to multi-view settings requires additional overhead modules or preprocessed data. To this end, we present CHROMM, a unified framework that jointly estimates cameras, scene point clouds, and human meshes from multi-person multi-view videos without relying on external modules or preprocessing. We integrate strong geometric and human priors from Pi3X and Multi-HMR into a single trainable neural network architecture, and introduce a scale adjustment module to solve the scale discrepancy between humans and the scene. We also introduce a multi-view fusion strategy to aggregate per-view estimates into a single representation at test-time. Finally, we propose a geometry-based multi-person association method, which is more robust than appearance-based approaches. Experiments on EMDB, RICH, EgoHumans, and EgoExo4D show that CHROMM achieves competitive performance in global human motion and multi-view pose estimation while running over 8x faster than prior optimization-based multi-view approaches. Project page: https://nstar1125.github.io/chromm.

Jungho Lee, Donghyeong Kim, Dogyoon Lee et al.

3D Gaussian Splatting (3DGS) has gained significant attention due to its high-quality novel view rendering, motivating research to address real-world challenges. A critical issue is the camera motion blur caused by movement during exposure, which hinders accurate 3D scene reconstruction. In this study, we propose CoMoGaussian, a Continuous Motion-Aware Gaussian Splatting that reconstructs precise 3D scenes from motion-blurred images while maintaining real-time rendering speed. Considering the complex motion patterns inherent in real-world camera movements, we predict continuous camera trajectories using neural ordinary differential equations (ODEs). To ensure accurate modeling, we employ rigid body transformations, preserving the shape and size of the object but rely on the discrete integration of sampled frames. To better approximate the continuous nature of motion blur, we introduce a continuous motion refinement (CMR) transformation that refines rigid transformations by incorporating additional learnable parameters. By revisiting fundamental camera theory and leveraging advanced neural ODE techniques, we achieve precise modeling of continuous camera trajectories, leading to improved reconstruction accuracy. Extensive experiments demonstrate state-of-the-art performance both quantitatively and qualitatively on benchmark datasets, which include a wide range of motion blur scenarios, from moderate to extreme blur.