Jongwoo Lim

Jongbin Ryu, Dongyoon Han, Jongwoo Lim

We introduce a novel architecture design that enhances expressiveness by incorporating multiple head classifiers (\ie, classification heads) instead of relying on channel expansion or additional building blocks. Our approach employs attention-based aggregation, utilizing pairwise feature similarity to enhance multiple lightweight heads with minimal resource overhead. We compute the Gramian matrices to reinforce class tokens in an attention layer for each head. This enables the heads to learn more discriminative representations, enhancing their aggregation capabilities. Furthermore, we propose a learning algorithm that encourages heads to complement each other by reducing correlation for aggregation. Our models eventually surpass state-of-the-art CNNs and ViTs regarding the accuracy-throughput trade-off on ImageNet-1K and deliver remarkable performance across various downstream tasks, such as COCO object instance segmentation, ADE20k semantic segmentation, and fine-grained visual classification datasets. The effectiveness of our framework is substantiated by practical experimental results and further underpinned by generalization error bound. We release the code publicly at: https://github.com/Lab-LVM/imagenet-models.

Gunwoo Jeon, Chaesong Park, Jongwoo Lim

Event cameras capture brightness changes asynchronously with microsecond resolution, yet existing optical flow methods fail to fully exploit this temporal continuity. Frame-based approaches impose artificial accumulation latency and suffer from domain overfitting, while model-based local methods operate statelessly, discarding temporal history between predictions and yielding inaccurate flows. We propose \textbf{LC-Flow}, the first temporally continuous, learning-based optical flow estimator that operates purely from local events. At its core, a Continuous Local Recurrent Network maintains persistent hidden states per spatial grid, incrementally accumulating temporal context as events arrive. Unlike frame-based methods constrained to fixed accumulation windows, and unlike stateless model-based methods that recompute motion from scratch at each step, LC-Flow produces sparse local flow estimates at arbitrary timestamps with full motion history. To address the inherent ambiguity of local observations, we jointly learn a confidence score that quantifies the reliability of each prediction, explicitly handling event sparsity and the aperture problem. This confidence serves a dual role: filtering unreliable estimates for downstream tasks such as visual odometry, and providing principled weights for a multi-scale confidence-guided aggregation that reconstructs globally consistent flow from the sparse local outputs. LC-Flow achieves state-of-the-art performance among local methods on both MVSEC and DSEC, while the confidence-guided aggregation establishes a new overall state-of-the-art on the MVSEC benchmark, surpassing heavy frame-based networks that rely on global spatial priors.

Jiyeop Kim, Jongwoo Lim

We present a novel approach for 3D indoor scene reconstruction that combines 3D Gaussian Splatting (3DGS) with mesh representations. We use meshes for the room layout of the indoor scene, such as walls, ceilings, and floors, while employing 3D Gaussians for other objects. This hybrid approach leverages the strengths of both representations, offering enhanced flexibility and ease of editing. However, joint training of meshes and 3D Gaussians is challenging because it is not clear which primitive should affect which part of the rendered image. Objects close to the room layout often struggle during training, particularly when the room layout is textureless, which can lead to incorrect optimizations and unnecessary 3D Gaussians. To overcome these challenges, we employ Segment Anything Model (SAM) to guide the selection of primitives. The SAM mask loss enforces each instance to be represented by either Gaussians or meshes, ensuring clear separation and stable training. Furthermore, we introduce an additional densification stage without resetting the opacity after the standard densification. This stage mitigates the degradation of image quality caused by a limited number of 3D Gaussians after the standard densification.

Chaehyeon Song, Jaeho Shin, Myung-Hwan Jeon et al.

In the literature, points and conics have been major features for camera geometric calibration. Although conics are more informative features than points, the loss of the conic property under distortion has critically limited the utility of conic features in camera calibration. Many existing approaches addressed conic-based calibration by ignoring distortion or introducing 3D spherical targets to circumvent this limitation. In this paper, we present a novel formulation for conic-based calibration using moments. Our derivation is based on the mathematical finding that the first moment can be estimated without bias even under distortion. This allows us to track moment changes during projection and distortion, ensuring the preservation of the first moment of the distorted conic. With an unbiased estimator, the circular patterns can be accurately detected at the sub-pixel level and can now be fully exploited for an entire calibration pipeline, resulting in significantly improved calibration. The entire code is readily available from https://github.com/ChaehyeonSong/discocal.

Chaerin Min, Sehyun Cha, Changhee Won et al.

Multi-view neural surface reconstruction has exhibited impressive results. However, a notable limitation is the prohibitively slow inference time when compared to traditional techniques, primarily attributed to the dense sampling, required to maintain the rendering quality. This paper introduces a novel approach that substantially reduces the number of samplings by incorporating the Truncated Signed Distance Field (TSDF) of the scene. While prior works have proposed importance sampling, their dependence on initial uniform samples over the entire space makes them unable to avoid performance degradation when trying to use less number of samples. In contrast, our method leverages the TSDF volume generated only by the trained views, and it proves to provide a reasonable bound on the sampling from upcoming novel views. As a result, we achieve high rendering quality by fully exploiting the continuous neural SDF estimation within the bounds given by the TSDF volume. Notably, our method is the first approach that can be robustly plug-and-play into a diverse array of neural surface field models, as long as they use the volume rendering technique. Our empirical results show an 11-fold increase in inference speed without compromising performance. The result videos are available at our project page: https://tsdf-sampling.github.io/

Cao Dinh Duc, Jongwoo Lim

Segmentation of planar regions from a single RGB image is a particularly important task in the perception of complex scenes. To utilize both visual and geometric properties in images, recent approaches often formulate the problem as a joint estimation of planar instances and dense depth through feature fusion mechanisms and geometric constraint losses. Despite promising results, these methods do not consider cross-task feature distillation and perform poorly in boundary regions. To overcome these limitations, we propose X-PDNet, a framework for the multitask learning of plane instance segmentation and depth estimation with improvements in the following two aspects. Firstly, we construct the cross-task distillation design which promotes early information sharing between dual-tasks for specific task improvements. Secondly, we highlight the current limitations of using the ground truth boundary to develop boundary regression loss, and propose a novel method that exploits depth information to support precise boundary region segmentation. Finally, we manually annotate more than 3000 images from Stanford 2D-3D-Semantics dataset and make available for evaluation of plane instance segmentation. Through the experiments, our proposed methods prove the advantages, outperforming the baseline with large improvement margins in the quantitative results on the ScanNet and the Stanford 2D-3D-S dataset, demonstrating the effectiveness of our proposals.

Chaehyeon Song, Dongjae Lee, Jongwoo Lim et al.

Camera calibration using planar targets has been widely favored, and two types of control points have been mainly considered as measurements: the corners of the checkerboard and the centroid of circles. Since a centroid is derived from numerous pixels, the circular pattern provides more precise measurements than the checkerboard. However, the existing projection model of circle centroids is biased under lens distortion, resulting in low performance. To surmount this limitation, we propose an unbiased projection model of the circular pattern and demonstrate its superior accuracy compared to the checkerboard. Complementing this, we introduce uncertainty into circular patterns to enhance calibration robustness and completeness. Defining centroid uncertainty improves the performance of calibration components, including pattern detection, optimization, and evaluation metrics. We also provide guidelines for performing good camera calibration based on the evaluation metric. The core concept of this approach is to model the boundary points of a two-dimensional shape as a Markov random field, considering its connectivity. The shape distribution is propagated to the centroid uncertainty through an appropriate shape representation based on the Green theorem. Consequently, the resulting framework achieves marked gains in calibration accuracy and robustness. The complete source code and demonstration video are available at https://github.com/chaehyeonsong/discocal.

Chaesong Park, Eunbin Seo, Jongwoo Lim

Accurate 3D lane detection from monocular images presents significant challenges due to depth ambiguity and imperfect ground modeling. Previous attempts to model the ground have often used a planar ground assumption with limited degrees of freedom, making them unsuitable for complex road environments with varying slopes. Our study introduces HeightLane, an innovative method that predicts a height map from monocular images by creating anchors based on a multi-slope assumption. This approach provides a detailed and accurate representation of the ground. HeightLane employs the predicted heightmap along with a deformable attention-based spatial feature transform framework to efficiently convert 2D image features into 3D bird's eye view (BEV) features, enhancing spatial understanding and lane structure recognition. Additionally, the heightmap is used for the positional encoding of BEV features, further improving their spatial accuracy. This explicit view transformation bridges the gap between front-view perceptions and spatially accurate BEV representations, significantly improving detection performance. To address the lack of the necessary ground truth (GT) height map in the original OpenLane dataset, we leverage the Waymo dataset and accumulate its LiDAR data to generate a height map for the drivable area of each scene. The GT heightmaps are used to train the heightmap extraction module from monocular images. Extensive experiments on the OpenLane validation set show that HeightLane achieves state-of-the-art performance in terms of F-score, highlighting its potential in real-world applications.

Mijeong Kim, Jongwoo Lim, Bohyung Han

Novel view synthesis of dynamic scenes is becoming important in various applications, including augmented and virtual reality. We propose a novel 4D Gaussian Splatting (4DGS) algorithm for dynamic scenes from casually recorded monocular videos. To overcome the overfitting problem of existing work for these real-world videos, we introduce an uncertainty-aware regularization that identifies uncertain regions with few observations and selectively imposes additional priors based on diffusion models and depth smoothness on such regions. This approach improves both the performance of novel view synthesis and the quality of training image reconstruction. We also identify the initialization problem of 4DGS in fast-moving dynamic regions, where the Structure from Motion (SfM) algorithm fails to provide reliable 3D landmarks. To initialize Gaussian primitives in such regions, we present a dynamic region densification method using the estimated depth maps and scene flow. Our experiments show that the proposed method improves the performance of 4DGS reconstruction from a video captured by a handheld monocular camera and also exhibits promising results in few-shot static scene reconstruction.

Sangheon Yang, Yeongin Yoon, Hong Mo Jung et al.

Traditional Visual Odometry (VO) and Visual Inertial Odometry (VIO) methods rely on a 'pose-centric' paradigm, which computes absolute camera poses from the local map thus requires large-scale landmark maintenance and continuous map optimization. This approach is computationally expensive, limiting their real-time performance on resource-constrained devices. To overcome these limitations, we introduce Sparse Motion Field Visual Odometry (SMF-VO), a lightweight, 'motion-centric' framework. Our approach directly estimates instantaneous linear and angular velocity from sparse optical flow, bypassing the need for explicit pose estimation or expensive landmark tracking. We also employed a generalized 3D ray-based motion field formulation that works accurately with various camera models, including wide-field-of-view lenses. SMF-VO demonstrates superior efficiency and competitive accuracy on benchmark datasets, achieving over 100 FPS on a Raspberry Pi 5 using only a CPU. Our work establishes a scalable and efficient alternative to conventional methods, making it highly suitable for mobile robotics and wearable devices.

Byeonggyeol Choi, Woojin Oh, Jongwoo Lim

Dexterous hand manipulation increasingly relies on large-scale motion datasets with precise hand-object trajectory data. However, existing resources such as DexYCB and HO3D are primarily optimized for visual alignment but often yield physically implausible interactions when replayed in physics simulators, including penetration, missed contact, and unstable grasps. We propose a simulation-in-the-loop refinement framework that converts these visually aligned trajectories into physically executable ones. Our core contribution is to formulate this as a tractable black-box optimization problem. We parameterize the hand's motion using a low-dimensional, spline-based representation built on sparse temporal keyframes. This allows us to use a powerful gradient-free optimizer, CMA-ES, to treat the high-fidelity physics engine as a black-box objective function. Our method finds motions that simultaneously maximize physical success (e.g., stable grasp and lift) while minimizing deviation from the original human demonstration. Compared to MANIPTRANS-recent transfer pipelines, our approach achieves lower hand and object pose errors during replay and more accurately recovers hand-object physical interactions. Our approach provides a general and scalable method for converting visual demonstrations into physically valid trajectories, enabling the generation of high-fidelity data crucial for robust policy learning.

Seongmin Jung, Seongho Choi, Gunwoo Jeon et al.

3D Visual Grounding (3DVG) is a critical bridge from vision-language perception to robotics, requiring both language understanding and 3D scene reasoning. Traditional supervised models leverage explicit 3D geometry but exhibit limited generalization, owing to the scarcity of 3D vision-language datasets and the limited reasoning capabilities compared to modern vision-language models (VLMs). We propose PanoGrounder, a generalizable 3DVG framework that couples multi-modal panoramic representation with pretrained 2D VLMs for strong vision-language reasoning. Panoramic renderings, augmented with 3D semantic and geometric features, serve as an intermediate representation between 2D and 3D, and offer two major benefits: (i) they can be directly fed to VLMs with minimal adaptation and (ii) they retain long-range object-to-object relations thanks to their 360-degree field of view. We devise a three-stage pipeline that places a compact set of panoramic viewpoints considering the scene layout and geometry, grounds a text query on each panoramic rendering with a VLM, and fuses per-view predictions into a single 3D bounding box via lifting. Our approach achieves state-of-the-art results on ScanRefer and Nr3D, and demonstrates superior generalization to unseen 3D datasets and text rephrasings.

Chaesong Park, Eunbin Seo, Jihyeon Hwang et al.

In this paper, we introduce SC-Lane, a novel slope-aware and temporally consistent heightmap estimation framework for 3D lane detection. Unlike previous approaches that rely on fixed slope anchors, SC-Lane adaptively determines the fusion of slope-specific height features, improving robustness to diverse road geometries. To achieve this, we propose a Slope-Aware Adaptive Feature module that dynamically predicts the appropriate weights from image cues for integrating multi-slope representations into a unified heightmap. Additionally, a Height Consistency Module enforces temporal coherence, ensuring stable and accurate height estimation across consecutive frames, which is crucial for real-world driving scenarios. To evaluate the effectiveness of SC-Lane, we employ three standardized metrics-Mean Absolute Error(MAE), Root Mean Squared Error (RMSE), and threshold-based accuracy-which, although common in surface and depth estimation, have been underutilized for road height assessment. Using the LiDAR-derived heightmap dataset introduced in prior work [20], we benchmark our method under these metrics, thereby establishing a rigorous standard for future comparisons. Extensive experiments on the OpenLane benchmark demonstrate that SC-Lane significantly improves both height estimation and 3D lane detection, achieving state-of-the-art performance with an F-score of 64.3%, outperforming existing methods by a notable margin. For detailed results and a demonstration video, please refer to our project page:https://parkchaesong.github.io/sclane/

Changhee Won, Hochang Seok, Zhaopeng Cui et al.

In this paper, we present an omnidirectional localization and dense mapping system for a wide-baseline multiview stereo setup with ultra-wide field-of-view (FOV) fisheye cameras, which has a 360 degrees coverage of stereo observations of the environment. For more practical and accurate reconstruction, we first introduce improved and light-weighted deep neural networks for the omnidirectional depth estimation, which are faster and more accurate than the existing networks. Second, we integrate our omnidirectional depth estimates into the visual odometry (VO) and add a loop closing module for global consistency. Using the estimated depth map, we reproject keypoints onto each other view, which leads to a better and more efficient feature matching process. Finally, we fuse the omnidirectional depth maps and the estimated rig poses into the truncated signed distance function (TSDF) volume to acquire a 3D map. We evaluate our method on synthetic datasets with ground-truth and real-world sequences of challenging environments, and the extensive experiments show that the proposed system generates excellent reconstruction results in both synthetic and real-world environments.

Jongbin Ryu, Jiun Bae, Jongwoo Lim

In this paper, we introduce a collaborative training algorithm of balanced random forests with convolutional neural networks for domain adaptation tasks. In real scenarios, most domain adaptation algorithms face the challenges from noisy, insufficient training data and open set categorization. In such cases, conventional methods suffer from overfitting and fail to successfully transfer the knowledge of the source to the target domain. To address these issues, the following two techniques are proposed. First, we introduce the optimized decision tree construction method with convolutional neural networks, in which the data at each node are split into equal sizes while maximizing the information gain. It generates balanced decision trees on deep features because of the even-split constraint, which contributes to enhanced discrimination power and reduced overfitting problem. Second, to tackle the domain misalignment problem, we propose the domain alignment loss which penalizes uneven splits of the source and target domain data. By collaboratively optimizing the information gain of the labeled source data as well as the entropy of unlabeled target data distributions, the proposed CoBRF algorithm achieves significantly better performance than the state-of-the-art methods.

Changhee Won, Jongbin Ryu, Jongwoo Lim

In this paper, we propose a novel end-to-end deep neural network model for omnidirectional depth estimation from a wide-baseline multi-view stereo setup. The images captured with ultra wide field-of-view (FOV) cameras on an omnidirectional rig are processed by the feature extraction module, and then the deep feature maps are warped onto the concentric spheres swept through all candidate depths using the calibrated camera parameters. The 3D encoder-decoder block takes the aligned feature volume to produce the omnidirectional depth estimate with regularization on uncertain regions utilizing the global context information. In addition, we present large-scale synthetic datasets for training and testing omnidirectional multi-view stereo algorithms. Our datasets consist of 11K ground-truth depth maps and 45K fisheye images in four orthogonal directions with various objects and environments. Experimental results show that the proposed method generates excellent results in both synthetic and real-world environments, and it outperforms the prior art and the omnidirectional versions of the state-of-the-art conventional stereo algorithms.

Hochang Seok, Jongwoo Lim

In this paper we propose a robust visual odometry system for a wide-baseline camera rig with wide field-of-view (FOV) fisheye lenses, which provides full omnidirectional stereo observations of the environment. For more robust and accurate ego-motion estimation we adds three components to the standard VO pipeline, 1) the hybrid projection model for improved feature matching, 2) multi-view P3P RANSAC algorithm for pose estimation, and 3) online update of rig extrinsic parameters. The hybrid projection model combines the perspective and cylindrical projection to maximize the overlap between views and minimize the image distortion that degrades feature matching performance. The multi-view P3P RANSAC algorithm extends the conventional P3P RANSAC to multi-view images so that all feature matches in all views are considered in the inlier counting for robust pose estimation. Finally the online extrinsic calibration is seamlessly integrated in the backend optimization framework so that the changes in camera poses due to shocks or vibrations can be corrected automatically. The proposed system is extensively evaluated with synthetic datasets with ground-truth and real sequences of highly dynamic environment, and its superior performance is demonstrated.

Changhee Won, Jongbin Ryu, Jongwoo Lim

Omnidirectional depth sensing has its advantage over the conventional stereo systems since it enables us to recognize the objects of interest in all directions without any blind regions. In this paper, we propose a novel wide-baseline omnidirectional stereo algorithm which computes the dense depth estimate from the fisheye images using a deep convolutional neural network. The capture system consists of multiple cameras mounted on a wide-baseline rig with ultrawide field of view (FOV) lenses, and we present the calibration algorithm for the extrinsic parameters based on the bundle adjustment. Instead of estimating depth maps from multiple sets of rectified images and stitching them, our approach directly generates one dense omnidirectional depth map with full 360-degree coverage at the rig global coordinate system. To this end, the proposed neural network is designed to output the cost volume from the warped images in the sphere sweeping method, and the final depth map is estimated by taking the minimum cost indices of the aggregated cost volume by SGM. For training the deep neural network and testing the entire system, realistic synthetic urban datasets are rendered using Blender. The experiments using the synthetic and real-world datasets show that our algorithm outperforms the conventional depth estimation methods and generate highly accurate depth maps.

Shun Zhang, Jia-Bin Huang, Jongwoo Lim et al.

Multi-face tracking in unconstrained videos is a challenging problem as faces of one person often appear drastically different in multiple shots due to significant variations in scale, pose, expression, illumination, and make-up. Existing multi-target tracking methods often use low-level features which are not sufficiently discriminative for identifying faces with such large appearance variations. In this paper, we tackle this problem by learning discriminative, video-specific face representations using convolutional neural networks (CNNs). Unlike existing CNN-based approaches which are only trained on large-scale face image datasets offline, we use the contextual constraints to generate a large number of training samples for a given video, and further adapt the pre-trained face CNN to specific videos using discovered training samples. Using these training samples, we optimize the embedding space so that the Euclidean distances correspond to a measure of semantic face similarity via minimizing a triplet loss function. With the learned discriminative features, we apply the hierarchical clustering algorithm to link tracklets across multiple shots to generate trajectories. We extensively evaluate the proposed algorithm on two sets of TV sitcoms and YouTube music videos, analyze the contribution of each component, and demonstrate significant performance improvement over existing techniques.

Sehyung Lee, Jongwoo Lim, Il Hong Suh

We present a novel feature matching algorithm that systematically utilizes the geometric properties of features such as position, scale, and orientation, in addition to the conventional descriptor vectors. In challenging scenes with the presence of repetitive patterns or with a large viewpoint change, it is hard to find the correct correspondences using feature descriptors only, since the descriptor distances of the correct matches may not be the least among the candidates due to appearance changes. Assuming that the layout of the nearby features does not changed much, we propose the bidirectional transfer measure to gauge the geometric consistency of a pair of feature correspondences. The feature matching problem is formulated as a Markov random field (MRF) which uses descriptor distances and relative geometric similarities together. The unmatched features are explicitly modeled in the MRF to minimize its negative impact. For speed and stability, instead of solving the MRF on the entire features at once, we start with a small set of confident feature matches, and then progressively search the candidates in nearby features and expand the MRF with them. Experimental comparisons show that the proposed algorithm finds better feature correspondences, i.e. more matches with higher inlier ratio, in many challenging scenes with much lower computational cost than the state-of-the-art algorithms.

Longyin Wen, Dawei Du, Zhaowei Cai et al.

In recent years, numerous effective multi-object tracking (MOT) methods are developed because of the wide range of applications. Existing performance evaluations of MOT methods usually separate the object tracking step from the object detection step by using the same fixed object detection results for comparisons. In this work, we perform a comprehensive quantitative study on the effects of object detection accuracy to the overall MOT performance, using the new large-scale University at Albany DETection and tRACking (UA-DETRAC) benchmark dataset. The UA-DETRAC benchmark dataset consists of 100 challenging video sequences captured from real-world traffic scenes (over 140,000 frames with rich annotations, including occlusion, weather, vehicle category, truncation, and vehicle bounding boxes) for object detection, object tracking and MOT system. We evaluate complete MOT systems constructed from combinations of state-of-the-art object detection and object tracking methods. Our analysis shows the complex effects of object detection accuracy on MOT system performance. Based on these observations, we propose new evaluation tools and metrics for MOT systems that consider both object detection and object tracking for comprehensive analysis.