Novel Category Discovery (NCD) methods for LiDAR Semantic Segmentation (LSS) assume that labeled and unlabeled points coexist in every scan and that all unlabeled points belong solely to novel categories. We formalize this into a more practical task, Generalized Category Discovery (GCD) for LSS, in which the labeled and unlabeled subsets are disjoint, and the unlabeled data contain a mixture of known and novel categories. Existing 2D GCD methods fail under this setting, struggling to distinguish the two groups in sparse, imbalanced LiDAR data. To address this limitation, we present a unified framework that (i) employs a learnable adaptive threshold to obtain point-wise anomaly scores to capture candidates, (ii) refines these candidates through a clustering-based filtering mechanism, and (iii) stabilizes training with a novel-feature queue that supplies reliable novel features even when a scene lacks novel categories. This explicit modeling of novel categories preserves segmentation quality for known classes while markedly improving discovery performance—a direction not explored in prior GCD or NCD methods. Extensive experiments on SemanticKITTI and nuScenes demonstrate that our approach consistently surpasses adapted baselines, establishing a strong benchmark for future work in open-world LSS.

Multiple Object Tracking (MOT) in dynamic and densely populated scenes presents significant challenges due to frequent occlusions, erratic object motion, and identity switches. While recent Transformer-based approaches have successfully leveraged global attention for object detection, most rely on temporal reasoning at the decoder level, leaving encoder-stage modeling underexplored. In this work, we propose an encoder-level temporal reasoning Transformer framework that embeds historical object trajectory information into the encoder stage via a tracklet memory. The encoder module, enhanced by Attention-by-Tracking, enriches visual features with temporal priors, while the decoder leverages Tracking-by-Attention to guide identity association using learned tracklet representations. To further improve temporal consistency and object localization, we introduce a gating-based temporal feature fusion mechanism that adaptively integrates multi-frame features based on cosine similarity. Our method is model-agnostic and can be applied to various Transformer-based MOT frameworks. When integrated into existing models such as TransTrack, MeMOTR, and MOTIP, it yields consistent performance improvements. Extensive experiments on DanceTrack and SportsMOT benchmarks demonstrate that our approach achieves superior tracking performance, including a HOTA score of 76.4 on SportsMOT. These results validate the effectiveness of encoder-level temporal integration and adaptive feature fusion for robust multi-object tracking in real-world scenarios.

Open-set driving in complex real-world environments requires reliable identification of out-of-distribution (OOD) objects to avoid overconfident predictions on unseen categories. However, the sparsity and limited semantic richness of LiDAR point clouds make separating known and unknown classes difficult. This work proposes DOODLE, a diffusion model–based OOD learning framework for open-set 3D semantic segmentation. DOODLE trains a diffusion model to reconstruct in-distribution semantic features; feature-level reconstruction discrepancies then serve as OOD evidence. The resulting OOD scores are used to enhance backbone semantic features, improving discrimination of unknown regions during segmentation. To further reduce false positives arising from nonuniform measurements, a density-aware post-processing (DAP) module incorporates spatial variation in LiDAR point density when refining OOD predictions. DOODLE integrates seamlessly with existing open-set models and does not constrain backbone design. Experiments on SemanticKITTI and nuScenes demonstrate state-of-the-art OOD performance. On SemanticKITTI, DOODLE improves area under the precision–recall curve (AUPR) by 1.85%p and area under the receiver operating characteristic (AUROC) by 1.29%p over prior methods. Ablation studies confirm complementary benefits from diffusion-based reconstruction and DAP. Code is available at https://github.com/chang9711/DOODLE.

Event cameras are dynamic vision sensors inspired by the biological retina, offering high dynamic range, high temporal resolution, and low power consumption. These qualities allow them to perceive 3D environments even in extreme conditions. Event data is continuously recorded over time, capturing pixel movements in detail. To leverage this temporal density, we introduce a temporal event stereo framework that continuously uses past information. The event stereo matching network is jointly trained with stereoscopic flow, which tracks pixel movements from stereo cameras. Instead of relying on optical flow ground truth, our method trains motion flows using disparity maps. The temporal aggregation of information via stereoscopic flow boosts stereo matching performance, achieving state-of-the-art results on MVSEC, DSEC, M3ED, and EVIMO2 datasets. Our method also demonstrates computational efficiency by stacking past data in a cascading manner.

With the success of the 3D deep learning models, various perception technologies for autonomous driving have been developed in the LiDAR domain. While these models perform well in the trained source domain, they struggle in unseen domains with a domain gap. In this paper, we propose a representation learning approach for domain generalization in LiDAR semantic segmentation, termed DGLSS++, which is designed to ensure robust performance in both the source domain and unseen domains despite training exclusively on the source domain. Our approach focuses on generalizing from a single source domain, addressing the domain shift caused by variations in LiDAR sensor configurations and scene distributions. To tackle both sparse-to-dense and dense-to-sparse generalization scenarios, we simulate unseen domains by generating sparsely and densely augmented domains. With the augmented domain, we introduce two constraints for generalizable representation learning: generalized masked sparsity invariant feature consistency (GMSIFC) and localized semantic correlation consistency (LSCC). GMSIFC aligns the internal sparse features of the source domain with those of the augmented domain at different sparsity, introducing a novel masking strategy to exclude voxel features associated with multiple inconsistent classes. For LSCC, class prototypes from spatially local regions are constrained to maintain similar correlations across all local regions, regardless of the scene or domain. In addition, we establish standardized training and evaluation protocols utilizing four real-world datasets and implement several baseline methods. Extensive experiments demonstrate our approach outperforms both UDA and DG baselines.

Detecting 3D objects in point clouds plays a crucial role in autonomous driving systems. Recently, advanced multimodal methods incorporating camera information have achieved notable performance. For a safe and effective autonomous driving system, algorithms that excel not only in accuracy but also in speed and low latency are essential. However, existing algorithms fail to meet these requirements due to the latency and bandwidth limitations of fixed frame rate sensors, e.g., LiDAR and camera. To address this limitation, we introduce asynchronous event cameras into 3D object detection for the first time. We leverage their high temporal resolution and low bandwidth to enable high-speed 3D object detection. Our method enables detection even during inter-frame intervals when synchronized data is unavailable, by retrieving previous 3D information through the event camera. Furthermore, we introduce the first event-based 3D object detection datasets, Ev-Waymo and DSEC-3DOD, both of which include ground truth 3D bounding boxes at 100 FPS, establishing the first benchmarks for event-based 3D detectors. The code and dataset are available at https://github.com/mickeykang16/Ev3DOD.

Pedestrian trajectory forecasting is crucial in various applications such as autonomous driving and mobile robot navigation. In such applications, camera-based perception enables the extraction of additional modalities (human pose, text) to enhance prediction accuracy. Indeed, we find that textual descriptions play a crucial role in integrating additional modalities into a unified understanding. However, online extraction of text requires the use of VLM, which may not be feasible for resource-constrained systems. To address this challenge, we propose a multimodal knowledge distillation framework: a student model with limited modality is distilled from a teacher model trained with full range of modalities. The comprehensive knowledge of a teacher model trained with trajectory, human pose, and text is distilled into a student model using only trajectory or human pose as a sole supplement. In doing so, we separately distill the core locomotion insights from intra-agent multi-modality and inter-agent interaction. Our generalizable framework is validated with two state-of-the-art models across three datasets on both ego-view (JRDB, SIT) and BEV-view (ETH/UCY) setups, utilizing both annotated and VLM-generated text captions. Distilled student models show consistent improvement in all prediction metrics for both full and instantaneous observations, improving up to ∼13%. The code is available at github.com/Jaewoo97/KDTF.

Instance segmentation traditionally relies on dense pixel-level annotations, making it costly and labor-intensive. To alleviate this burden, weakly supervised instance segmentation utilizes cost-effective weak labels, such as image-level tags, points, and bounding boxes. However, existing approaches typically focus on a single type of weak label, overlooking the cost-efficiency potential of combining multiple types. In this paper, we introduce WISH, a novel heterogeneous framework for weakly supervised instance segmentation that integrates diverse weak label types within a single model. WISH unifies heterogeneous labels by leveraging SAM’s prompt latent space through a multi-stage matching strategy, effectively compensating for the lack of spatial information in class tags. Extensive experiments on Pascal VOC and COCO demonstrate that our framework not only surpasses existing homogeneous weak supervision methods but also achieves superior results in heterogeneous settings with equivalent annotation costs.

We introduce Any6D, a model-free framework for 6D object pose estimation that requires only a single RGB-D anchor image to estimate both the 6D pose and size of unknown objects in novel scenes. Unlike existing methods that rely on textured 3D models or multiple viewpoints, Any6D leverages a joint object alignment process to enhance 2D-3D alignment and metric scale estimation for improved pose accuracy. Our approach integrates a render-and-compare strategy to generate and refine pose hypotheses, enabling robust performance in scenarios with occlusions, non-overlapping views, diverse lighting conditions, and large cross-environment variations. We evaluate our method on five challenging datasets: REAL275, ToyotaLight, HO3D, YCBINEOAT, and LM-O, demonstrating its effectiveness in significantly outperforming state-of-the-art methods for novel object pose estimation. Project page: https://taeyeop.com/any6d

Generalizing neural networks to unseen target domains is a significant challenge in real-world deployments. Test-time training (TTT) addresses this by using an auxiliary self-supervised task to reduce the domain gap caused by distribution shifts between the source and target. However, we find that when models are required to perform multiple tasks under domain shifts, conventional TTT methods suffer from unsynchronized task behavior, where the adaptation steps needed for optimal performance in one task may not align with the requirements of other tasks. To address this, we propose a novel TTT approach called Synchronizing Tasks for Test-time Training (S4T), which enables the concurrent handling of multiple tasks. The core idea behind S4T is that predicting task relations across domain shifts is key to synchronizing tasks during test time. To validate our approach, we apply S4T to conventional multi-task benchmarks, integrating it with traditional TTT protocols. Our empirical results show that S4T outperforms state-of-the-art TTT methods across various benchmarks.

Multi-Task Learning (MTL) enables multiple tasks to be learned within a shared network, but differences in objectives across tasks can cause negative transfer, where the learning of one task degrades another task's performance. While pre-trained transformers significantly improve MTL performance, their fixed network capacity and rigid structure limit adaptability. Previous dynamic network architectures attempt to address this but are inefficient as they directly convert shared parameters into task-specific ones. We propose Dynamic Token Modulation and Expansion (DTME-MTL), a framework applicable to any transformer-based MTL architecture. DTME-MTL enhances adaptability and reduces overfitting by identifying gradient conflicts in token space and applying adaptive solutions based on conflict type. Unlike prior methods that mitigate negative transfer by duplicating network parameters, DTME-MTL operates entirely in token space, enabling efficient adaptation without excessive parameter growth. Extensive experiments demonstrate that DTME-MTL consistently improves multi-task performance with minimal computational overhead, offering a scalable and effective solution for enhancing transformer-based MTL models.

Adverse weather conditions cause diverse and complex degradation patterns, driving the development of All-in-One (AiO) models. However, recent AiO solutions still struggle to capture diverse degradations, since global filtering methods like direct operations on the frequency domain fail to handle highly variable and localized distortions. To address these issue, we propose Spectral-based Spatial Grouping Transformer (SSGformer), a novel approach that leverages spectral decomposition and group-wise attention for multi-weather image restoration. SSGformer decomposes images into high-frequency edge features using conventional edge detection and low-frequency information via Singular Value Decomposition. We utilize multi-head linear attention to effectively model the relationship between these features. The fused features are integrated with the input to generate a grouping-mask that clusters regions based on the spatial similarity and image texture. To fully leverage this mask, we introduce a group-wise attention mechanism, enabling robust adverse weather removal and ensuring consistent performance across diverse weather conditions. We also propose a Spatial Grouping Transformer Block that uses both channel attention and spatial attention, effectively balancing feature-wise relationships and spatial dependencies. Extensive experiments show the superiority of our approach, validating its effectiveness in handling the varied and intricate adverse weather degradations.

3D object detection is essential for autonomous systems, enabling precise localization and dimension estimation. While LiDAR and RGB cameras are widely used, their fixed frame rates create perception gaps in high-speed scenarios. Event cameras, with their asynchronous nature and high temporal resolution, offer a solution by capturing motion continuously. The recent approach, which integrates event cameras with conventional sensors for continuous-time detection, struggles in fast-motion scenarios due to its dependency on synchronized sensors. We propose a novel stereo 3D object detection framework that relies solely on event cameras, eliminating the need for conventional 3D sensors. To compensate for the lack of semantic and geometric information in event data, we introduce a dual filter mechanism that extracts both. Additionally, we enhance regression by aligning bounding boxes with object-centric information. Experiments show that our method outperforms prior approaches in dynamic environments, demonstrating the potential of event cameras for robust, continuous-time 3D perception. The code is available at https://github.com/mickeykang16/Ev-Stereo3D.

Motion planning is a crucial component of autonomous robot driving. While various trajectory datasets exist, effectively utilizing them for a target domain remains challenging due to differences in agent interactions and environmental characteristics. Conventional approaches, such as domain adaptation or ensemble learning, leverage multiple source datasets but suffer from domain imbalance, catastrophic forgetting, and high computational costs. To address these challenges, we propose Interaction-Merged Motion Planning (IMMP), a novel approach that leverages parameter checkpoints trained on different domains during adaptation to the target domain. IMMP follows a two-step process: pre-merging to capture agent behaviors and interactions, sufficiently extracting diverse information from the source domain, followed by merging to construct an adaptable model that efficiently transfers diverse interactions to the target domain. Our method is evaluated on various planning benchmarks and models, demonstrating superior performance compared to conventional approaches.

Multi-task learning enables the acquisition of task-generic knowledge by training multiple tasks within a unified architecture. However, training all tasks together in a single architecture can lead to performance degradation, known as negative transfer, which is a main concern in multi-task learning. Previous works have addressed this issue by optimizing the multi-task network through gradient manipulation or weighted loss adjustments. However, their optimization strategy focuses on addressing task imbalance in shared parameters, neglecting the learning of task-specific parameters. As a result, they show limitations in mitigating negative transfer, since the learning of shared space and task-specific information influences each other during optimization. To address this, we propose a different approach to enhance multi-task performance by selectively grouping tasks and updating them for each batch during optimization. We introduce an algorithm that adaptively determines how to effectively group tasks and update them during the learning process. To track inter-task relations and optimize multi-task networks simultaneously, we propose proximal inter-task affinity, which can be measured during the optimization process. We provide a theoretical analysis on how dividing tasks into multiple groups and updating them sequentially significantly affects multi-task performance by enhancing the learning of task-specific parameters. Our methods substantially outperform previous multi-task optimization approaches and are scalable to different architectures and various numbers of tasks.

Safe and effective motion planning is crucial for autonomous robots. Diffusion models excel at capturing complex agent interactions, a fundamental aspect of decision-making in dynamic environments. Recent studies have successfully applied diffusion models to motion planning, demonstrating their competence in handling complex scenarios and accurately predicting multi-modal future trajectories. Despite their effectiveness, diffusion models have limitations in training objectives, as they approximate data distributions rather than explicitly capturing the underlying decision-making dynamics. However, the crux of motion planning lies in non-differentiable downstream objectives, such as safety (collision avoidance) and effectiveness (goal-reaching), which conventional learning algorithms cannot directly optimize. In this paper, we propose a reinforcement learning-based training scheme for diffusion motion planning models, enabling them to effectively learn non-differentiable objectives that explicitly measure safety and effectiveness. Specifically, we introduce a reward-weighted dynamic thresholding algorithm to shape a dense reward signal, facilitating more effective training and outperforming models trained with differentiable objectives. State-of-the-art performance on pedestrian datasets (CrowdNav, ETH-UCY) compared to various baselines demonstrates the versatility of our approach for safe and effective motion planning.

End-to-end autonomous driving (E2E-AD) has emerged as a promising paradigm that unifies perception, prediction, and planning into a holistic, data-driven framework. However, achieving robustness to varying camera viewpoints, a common real-world challenge due to diverse vehicle configurations, remains an open problem. In this work, we propose VR-Drive, a novel E2E-AD framework that addresses viewpoint generalization by jointly learning 3D scene reconstruction as an auxiliary task to enable planning-aware view synthesis. Unlike prior scene-specific synthesis approaches, VR-Drive adopts a feed-forward inference strategy that supports online training-time augmentation from sparse views without additional annotations. To further improve viewpoint consistency, we introduce a viewpoint-mixed memory bank that facilitates temporal interaction across multiple viewpoints and a viewpoint-consistent distillation strategy that transfers knowledge from original to synthesized views. Trained in a fully end-to-end manner, VR-Drive effectively mitigates synthesis-induced noise and improves planning under viewpoint shifts. In addition, we release a new benchmark dataset to evaluate E2E-AD performance under novel camera viewpoints, enabling comprehensive analysis. Our results demonstrate that VR-Drive is a scalable and robust solution for the real-world deployment of end-to-end autonomous driving systems.

Event cameras deliver asynchronous pixel intensity changes, which result in sparse event data that offers the advantages of high temporal resolution. These high temporal characteristics make researchers naturally incorporate event cameras into video frame interpolation (VFI) and video super-resolution (VSR). In this paper, we make the first attempt to solve the space-time video super-resolution (STVSR) task effectively, addressing both VFI and VSR simultaneously, by leveraging temporally dense events. STVSR aims to generate intermediate high-resolution (HR) videos between consecutive low-resolution (LR) frames. To fully exploit the high temporal frequency of events for STVSR, we focus on temporal alignment in two stages, at low-resolution and after up-sampling in high-resolution. In temporal alignment at low-resolution, to upsample spatial dimensions effectively, we leverage high temporal features to preserve spatial context. On the other hand, for temporal alignment at the high-resolution stage, we employ a deformable sampling process from events to achieve accurate alignment with forward and backward directions. In addition, we provide the SuperREST dataset, which features high-frequency details and complex motion in an RGB-Event setup. Experimental results on several datasets demonstrate that our method achieves a significant performance gain on STVSR tasks with low computational cost. Our codes and datasets are available at h t t ps: //github.com/Chohoonhee/ESTNet.

This paper proposes an algorithm for synthesizing novel views under few-shot setup. The main concept is to develop a stable surface regularization technique called Annealing Signed Distance Function (ASDF), which anneals the surface in a coarse-to-fine manner to accelerate convergence speed. We observe that the Eikonal loss – which is a widely known geometric regularization – requires dense training signal to shape different level-sets of SDF, leading to low-fidelity results under few-shot training. In contrast, the proposed surface regularization successfully reconstructs scenes and produce high-fidelity geometry with stable training. Our method is further accelerated by utilizing grid representation and monocular geometric priors. Finally, the proposed approach is up to 45 times faster than existing few-shot novel view synthesis methods, and it produces comparable results in the ScanNet dataset and NeRF-Real dataset.

Multi-agent trajectory prediction is crucial for various practical applications, spurring the construction of many large-scale trajectory datasets, including vehicles and pedestrians. However, discrepancies exist among datasets due to external factors and data acquisition strategies. External factors include geographical differences and driving styles, while data acquisition strategies include data acquisition rate, history/prediction length, and detector/tracker error. Consequently, the proficient performance of models trained on large-scale datasets has limited transferability on other small-size datasets, bounding the utilization of existing large-scale datasets. To address this limitation, we propose a method based on continuous and stochastic representations of Neural Stochastic Differential Equations (NSDE) for alleviating discrepancies due to data acquisition strategy. We utilize the benefits of continuous representation for handling arbitrary time steps and the use of stochastic representation for handling detector/tracker errors. Additionally, we propose a dataset-specific diffusion network and its training framework to handle dataset-specific detection/tracking errors. The effectiveness of our method is validated against state-of-the-art trajectory prediction models on the popular benchmark datasets: nuScenes, Argoverse, Lyft, INTERACTION, and Waymo Open Motion Dataset (WOMD). Improvement in performance gain on various source and target dataset configurations shows the generalized competence of our approach in addressing cross-dataset discrepancies.

Deep neural networks are known to be vulnerable to security risks due to the inherent transferable nature of adversarial examples. Despite the success of recent generative model-based attacks demonstrating strong transferability, it still remains a challenge to design an efficient attack strategy in a real-world strict black-box setting, where both the target domain and model architectures are unknown. In this paper, we seek to explore a feature contrastive approach in the frequency domain to generate adversarial examples that are robust in both cross-domain and cross-model settings. With that goal in mind, we propose two modules that are only employed during the training phase: a Frequency-Aware Domain Randomization (FADR) module to randomize domain-variant low- and high-range frequency components and a Frequency-Augmented Contrastive Learning (FACL) module to effectively separate domain-invariant mid-frequency features of clean and perturbed image. We demonstrate strong transferability of our generated adversarial perturbations through extensive cross-domain and cross-model experiments, while keeping the inference time complexity.

Video deblurring aims to restore sharp frames from blurred video clips. Despite notable progress in video deblurring works, it is still a challenging problem because of the loss of motion information during the duration of the exposure time. Since event cameras can capture clear motion information asynchronously with high temporal resolution, several works exploit the event camera for deblurring as they can provide abundant motion information. However, despite these approaches, there were few cases of actively exploiting the long-range temporal dependency of videos. To tackle these deficiencies, we present an event-based video deblurring framework by actively utilizing temporal information from videos. To be specific, we first introduce a frequency-based cross-modal feature enhancement module. Second, we propose event-guided video alignment modules by considering the valuable characteristics of the event and videos. In addition, we designed a hybrid camera system to collect the first real-world event-based video deblurring dataset. For the first time, we build a dataset containing synchronized high-resolution real-world blurred videos and corresponding sharp videos and event streams. Experimental results validate that our frameworks significantly outperform the state-of-the-art frame-based and event-based deblurring works in the various datasets. The project pages are available at https://sites.google.com/view/fevd-cvpr2024.

Detecting objects in 3D under various (normal and adverse) weather conditions is essential for safe autonomous driving systems. Recent approaches have focused on employing weather-insensitive 4D radar sensors and leveraging them with other modalities, such as LiDAR. However, they fuse multi-modal information without considering the sensor characteristics and weather conditions, and lose some height information which could be useful for localizing 3D objects. In this paper, we propose a novel framework for robust LiDAR and 4D radar-based 3D object detection. Specifically, we propose a 3D-LRF module that considers the distinct patterns they exhibit in 3D space (e.g., precise 3D mapping of LiDAR and wide-range, weather-insensitive measurement of 4D radar) and extract fusion features based on their 3D spatial relationship. Then, our weather-conditional radar-flow gating network modulates the information flow of fusion features depending on weather conditions, and obtains enhanced feature that effectively incorporates the strength of two domains under various weather conditions. The extensive experiments demonstrate that our model achieves SoTA performance for 3D object detection under various weather conditions.

Video Frame Interpolation (VFI), which aims at gener-ating high-frame-rate videos from low-frame-rate inputs, is a highly challenging task. The emergence of bio-inspired sensors known as event cameras, which boast microsecond-level temporal resolution, has ushered in a transformative era for VFI. Nonetheless, the application of event-based VFI techniques in domains with distinct environments from the training data can be problematic. This is mainly because event camera data distribution can undergo substan-tial variations based on camera settings and scene conditions, presenting challenges for effective adaptation. In this paper, we propose a test-time adaptation method for event-based VFI to address the gap between the source and target domains. Our approach enables sequential learning in an online manner on the target domain, which only provides low-frame-rate videos. We present an approach that lever-ages confident pixels as pseudo ground-truths, enabling stable and accurate online learning from low-frame-rate videos. Furthermore, to prevent overfitting during the con-tinuous online process where the same scene is encountered repeatedly, we propose a method of blending historical sam-ples with current scenes. Extensive experiments validate the effectiveness of our method, both in cross-domain and con-tinuous domain shifting setups. The code is available at https://github.com/Chohoonhee/TTA-EVF.

Human pose forecasting garners attention for its diverse applications. However, challenges in modeling the multi-modal nature of human motion and intricate interactions among agents persist, particularly with longer timescales and more agents. In this paper, we propose an interaction-aware trajectory-conditioned long-term multiagent human pose forecasting model, utilizing a coarse-to-fine prediction approach: multi-modal global trajectories are initially forecasted, followed by respective local pose forecasts conditioned on each mode. In doing so, our Trajectory2Pose model introduces a graph-based agent-wise interaction module for a reciprocal forecast of local motion-conditioned global trajectory and trajectory-conditioned local pose. Our model effectively handles the multi-modality of human motion and the complexity of long-term multiagent interactions, improving performance in complex environments. Furthermore, we address the lack of long-term (6s+) multiagent (5 +) datasets by constructing a new dataset from real-world images and 2D annotations, enabling a comprehensive evaluation of our proposed model. State-of-the-art prediction performance on both complex and simpler datasets confirms the generalized effectiveness of our method. The code is available at https://github.com/Jaewoo97/T2P.

The goal of multi-task learning is to learn diverse tasks within a single unified network. As each task has its own unique objective function, conflicts emerge during training, resulting in negative transfer among them. Earlier research identified these conflicting gradients in shared parameters between tasks and attempted to realign them in the same direction. However, we prove that such optimization strategies lead to sub-optimal Pareto solutions due to their inability to accurately determine the individual contributions of each parameter across various tasks. In this paper, we propose the concept of task priority to evaluate parameter contributions across different tasks. To learn task priority, we identify the type of connections related to links between parameters influenced by task-specific losses during backpropagation. The strength of connections is gauged by the magnitude of parameters to determine task priority. Based on these, we present a new method named connection strength-based optimization for multi-task learning which consists of two phases. The first phase learns the task priority within the network, while the second phase modifies the gradients while upholding this priority. This ultimately leads to finding new Pareto optimal solutions for multiple tasks. Through extensive experiments, we show that our approach greatly enhances multi-task performance in comparison to earlier gradient manipulation methods.

Manual annotation of every point in a point cloud is a costly and labor-intensive process. While weakly super-vised point cloud semantic segmentation (WSPCSS) with sparse annotation shows promise, the limited information from initial sparse labels can place an upper bound on performance. As a new research direction for WSPCSS, we propose a novel Region Exploration via Artificial Labeling (REAL) framework. It leverages a foundational image model as an artificial oracle within the active learning context, eliminating the need for manual annotation by a human oracle. To integrate the 2D model into the 3D domain, we first introduce a Projection-based Point-to-Segment (PP2S) module, designed to enable prompt segmentation of 3D data without additional training. The REAL framework samples query points based on model predictions and requests annotations from PP2S, dynamically refining labels and improving model training. Furthermore, to overcome several challenges of employing an artificial model as an oracle, we formulate effective query sampling and label updating strategies. Our comprehensive experiments and comparisons demonstrate that the REAL framework significantly outperforms existing methods across various benchmarks. The code is available at https://github.com/jihun1998/AO.

Weakly Supervised Semantic Segmentation (WSSS) aims to learn the concept of segmentation using image-level class labels. Recent WSSS works have shown promising results by using the Segment Anything Model (SAM), a foundation model for segmentation, during the inference phase. However, we observe that these methods can still be vulnerable to the noise of class activation maps (CAMs) serving as initial seeds. As a remedy, this paper introduces From-SAM-to-CAMs (S2C), a novel WSSS framework that directly transfers the knowledge of SAM to the classifier during the training process, enhancing the quality of CAMs it-self. S2C comprises SAM-segment Contrasting (SSC) and a CAM-based prompting module (CPM), which exploit SAM at the feature and logit levels, respectively. SSC performs prototype-based contrasting using SAM's automatic segmentation results. It constrains each feature to be close to the prototype of its segment and distant from prototypes of the others. Meanwhile, CPM extracts prompts from the CAM of each class and uses them to generate classspecific segmentation masks through SAM. The masks are aggregated into unified self-supervision based on the confidence score, designed to consider the reliability of both SAM and CAMs. S2C achieves a new state-of-the-art performance across all benchmarks, outperforming existing studies by significant margins. The code is available at https://github.com/sangrockEG/S2C.

Trajectory prediction is a challenging problem that requires considering interactions among multiple actors and the surrounding environment. While data-driven approaches have been used to address this complex problem, they suffer from unreliable predictions under distribution shifts during test time. Accordingly, several online learning methods have been proposed using regression loss from the ground truth of observed data leveraging the auto-labeling nature of trajectory prediction task. We mainly tackle the following two issues. First, previous works underfit and overfit as they only optimize the last layer of motion decoder. To this end, we employ the masked autoencoder (MAE) for representation learning to encourage complex interaction modeling in shifted test distribution for updating deeper layers. Second, utilizing the sequential nature of driving data, we propose an actor-specific token memory that enables the test-time learning of actor-wise motion characteristics. Our proposed method has been validated across various challenging cross-dataset distribution shift scenarios including nuScenes, Lyft, Waymo, and Interaction. Our method surpasses the performance of existing state-of-the-art online learning methods in terms of both prediction accuracy and computational efficiency. The code is available at https://github.com/daeheepark/T4P.

Weakly Supervised Semantic Segmentation (WSSS) re-lies on Class Activation Maps (CAMs) to extract spatial information from image-level labels. With the success of Vision Transformer (ViT), the migration of ViT is actively conducted in WSSS. This work proposes a novel WSSS framework with Class Token Infusion (CTI). By infusing the class tokens from images, we guide class tokens to pos-sess class-specific distinct characteristics and global-local consistency. For this, we devise two kinds of token infusion: 1) Intra-image Class Token Infusion (I-CTI) and 2) Cross-image Class Token Infusion (C-CTI). In I-CTI, we infuse the class tokens from the same but differently augmented images and thus make CAMs consistent among var-ious deformations (i.e. view, color). In C-CTI, by infusing the class tokens from the other images and imposing the resulting CAMs to be similar, it learns class-specific distinct characteristics. Besides the CTI, we bring the background (BG) concept into ViT with the BG token to reduce the false positive activation of CAMs. We demonstrate the effectiveness of our method on PASCAL VOC 2012 and MS COCO 2014 datasets, achieving state-of-the-art results in weakly supervised semantic segmentation. The code is available at https://github.com/yoon307/CTI.

Multi-person pose estimation and tracking have been actively researched by the computer vision community due to their practical applicability. However, existing human pose estimation and tracking datasets have only been successful in typical scenarios, such as those without motion blur or with well-lit conditions. These RGB-based datasets are limited to learning under extreme motion blur situations or poor lighting conditions, making them inherently vulnerable to such scenarios.As a promising solution, bio-inspired event cameras exhibit robustness in extreme scenarios due to their high dynamic range and micro-second level temporal resolution. Therefore, in this paper, we introduce a new hybrid dataset encompassing both RGB and event data for human pose estimation and tracking in two extreme scenarios: low-light and motion blur environments. The proposed Event-guided Human Pose Estimation and Tracking in eXtreme Conditions (EHPT-XC) dataset covers cases of motion blur caused by dynamic objects and low-light conditions individually as well as both simultaneously. With EHPT-XC, we aim to inspire researchers to tackle pose estimation and tracking in extreme conditions by leveraging the advantageous of the event camera. Project pages are available at https://github.com/Chohoonhee/EHPT-XC.

Semantic Scene Completion (SSC) aims to perform geometric completion and semantic segmentation simultaneously. Despite the promising results achieved by existing studies, the inherently ill-posed nature of the task presents significant challenges in diverse driving scenarios. This paper introduces TALoS, a novel test-time adaptation approach for SSC that excavates the information available in driving environments. Specifically, we focus on that observations made at a certain moment can serve as Ground Truth (GT) for scene completion at another moment. Given the characteristics of the LiDAR sensor, an observation of an object at a certain location confirms both 1) the occupation of that location and 2) the absence of obstacles along the line of sight from the LiDAR to that point. TALoS utilizes these observations to obtain self-supervision about occupancy and emptiness, guiding the model to adapt to the scene in test time. In a similar manner, we aggregate reliable SSC predictions among multiple moments and leverage them as semantic pseudo-GT for adaptation. Further, to leverage future observations that are not accessible at the current time, we present a dual optimization scheme using the model in which the update is delayed until the future observation is available. Evaluations on the SemanticKITTI validation and test sets demonstrate that TALoS significantly improves the performance of the pre-trained SSC model.

Video Panoptic Segmentation (VPS) aims to achieve comprehensive pixel-level scene understanding by segmenting all pixels and associating objects in a video. Current solutions can be categorized into online and near-online approaches. Evolving over the time, each category has its own specialized designs, making it nontrivial to adapt models between different categories. To alleviate the discrepancy, in this work, we propose a unified approach for online and near-online VPS. The meta architecture of the proposed Video-kMaX consists of two components: within-clip segmenter (for clip-level segmentation) and cross-clip associater (for association beyond clips). We propose clip-kMaX (clip k-means mask transformer) and LA-MB (location-aware memory buffer) to instantiate the segmenter and associater, respectively. Our general formulation includes the online scenario as a special case by adopting clip length of one. Without bells and whistles, Video-kMaX sets a new state-of-the-art on KITTI-STEP and VIPSeg for video panoptic segmentation Code and models are available at this link.

Existing image stitching approaches based on global or local homography estimation are not free from the parallax problem and suffer from undesired artifacts. In this paper, instead of relying on the homography-based warp, we propose a novel deep image stitching framework exploiting the pixel-wise warp field to handle the large-parallax problem. The proposed deep image stitching framework consists of a Pixel-wise Warping Module (PWM) and a Stitched Image Generating Module (SIGMo). For PWM, we obtain pixel-wise warp in a similar manner as estimating an optical flow (OF). In the stitching scenario, the input images usually include non-overlap (NOV) regions of which warp cannot be directly estimated, unlike the overlap (OV) regions. To help the PWM predict a reasonable warp on the NOV region, we impose two geometrical constraints: an epipolar loss and a line-preservation loss. With the obtained warp field, we relocate the pixels of the target image using forward warping. Finally, the SIGMo is trained by the proposed multi-branch training framework to generate a stitched image from a reference image and a warped target image. For training and evaluating the proposed framework, we build and publish a novel dataset including image pairs with corresponding pixel-wise ground truth warp and stitched result images. We show that the results of the proposed framework are quantitatively and qualitatively superior to those of the conventional methods.

Recently, event-based stereo matching has been studied due to its robustness in poor light conditions. However, existing event-based stereo networks suffer severe performance degradation when domains shift. Unsupervised domain adaptation (UDA) aims at resolving this problem without using the target domain ground-truth. However, traditional UDA still needs the input event data with ground- truth in the source domain, which is more challenging and costly to obtain than image data. To tackle this issue, we propose a novel unsupervised domain Adaptive Dense Event Stereo (ADES), which resolves gaps between the different domains and input modalities. The proposed ADES framework adapts event-based stereo networks from abundant image datasets with ground-truth on the source domain to event datasets without ground-truth on the target domain, which is a more practical setup. First, we propose a self-supervision module that trains the network on the target domain through image reconstruction, while an artifact prediction network trained on the source domain assists in removing intermittent artifacts in the reconstructed image. Secondly, we utilize the feature-level normalization scheme to align the extracted features along the epipolar line. Finally, we present the motion-invariant consistency module to impose the consistent output between the perturbed motion. Our experiments demonstrate that our approach achieves remarkable results in the adaptation ability of event-based stereo matching from the image domain.

Video Frame Interpolation (VFI) aims to generate intermediate video frames between consecutive input frames. Since the event cameras are bio-inspired sensors that only encode brightness changes with a micro-second temporal resolution, several works utilized the event camera to enhance the performance of VFI. However, existing methods estimate bidirectional inter-frame motion fields with only events or approximations, which can not consider the complex motion in real-world scenarios. In this paper, we propose a novel event-based VFI framework with crossmodal asymmetric bidirectional motion field estimation. In detail, our EIF-BiOFNet utilizes each valuable characteristic of the events and images for direct estimation of inter-frame motion fields without any approximation methods. Moreover, we develop an interactive attention-based frame synthesis network to efficiently leverage the complementary warping-based and synthesis-based features. Finally, we build a large-scale event-based VFI dataset, ERF-X170FPS, with a high frame rate, extreme motion, and dynamic textures to overcome the limitations of previous event-based VFI datasets. Extensive experimental results validate that our method shows significant performance improvement over the state-of-the-art VFI methods on various datasets. Our project pages are available at: https://github.com/intelpro/CBMNet

With the success of the 3D deep learning models, various perception technologies for autonomous driving have been developed in the LiDAR domain. While these models perform well in the trained source domain, they struggle in unseen domains with a domain gap. In this paper, we propose a single domain generalization method for LiDAR semantic segmentation (DGLSS) that aims to ensure good performance not only in the source domain but also in the unseen domain by learning only on the source domain. We mainly focus on generalizing from a dense source domain and target the domain shift from different LiDAR sensor configurations and scene distributions. To this end, we augment the domain to simulate the unseen domains by randomly subsampling the LiDAR scans. With the augmented domain, we introduce two constraints for generalizable representation learning: sparsity invariant feature consistency (SIFC) and semantic correlation consistency (SCC). The SIFC aligns sparse internal features of the source domain with the augmented domain based on the feature affinity. For SCC, we constrain the correlation between class prototypes to be similar for every LiDAR scan. We also establish a standardized training and evaluation setting for DGLSS. With the proposed evaluation setting, our method showed improved performance in the unseen domains compared to other baselines. Even without access to the target domain, our method performed better than the domain adaptation method. The code is available at https://github.com/gzgzys9887/DGLSS.

In Weakly Supervised Semantic Segmentation (WSSS), Class Activation Maps (CAMs) usually 1) do not cover the whole object and 2) be activated on irrelevant regions. To address the issues, we propose a novel WSSS framework via adversarial learning of a classifier and an image reconstructor. When an image is perfectly decomposed into class-wise segments, information (i.e., color or texture) of a single segment could not be inferred from the other segments. Therefore, inferability between the segments can represent the preciseness of segmentation. We quantify the inferability as a reconstruction quality of one segment from the other segments. If one segment could be reconstructed from the others, then the segment would be imprecise. To bring this idea into WSSS, we simultaneously train two models: a classifier generating CAMs that decompose an image into segments and a reconstructor that measures the inferability between the segments. As in GANs, while being alternatively trained in an adversarial manner, two networks provide positive feedback to each other. We verify the superiority of the proposed framework with extensive ablation studies. Our method achieves new state-of-the-art performances on both PAS-CAL VOC 2012 and MS COCO 2014. The code is available at https://github.com/sangrockEG/ACR.

Test-time adaptation methods have been gaining attention recently as a practical solution for addressing source-to-target domain gaps by gradually updating the model without requiring labels on the target data. In this paper, we propose a method of test-time adaptation for category-level object pose estimation called TTA-COPE. We design a pose ensemble approach with a self-training loss using pose-aware confidence. Unlike previous unsupervised domain adaptation methods for category-level object pose estimation, our approach processes the test data in a sequential, online manner, and it does not require access to the source domain at runtime. Extensive experimental results demonstrate that the proposed pose ensemble and the self-training loss improve category-level object pose performance during test time under both semi-supervised and unsupervised settings.

Novel View Synthesis (NVS) aims at synthesizing an image from an arbitrary viewpoint using multi-view images and camera poses. Among the methods for NVS, Neural Radiance Fields (NeRF) is capable of NVS for an arbitrary resolution as it learns a continuous volumetric representation. However, radiance fields rely heavily on the spectral characteristics of coordinate-based networks. Thus, there is a limit to improving the performance of high-resolution novel view synthesis (HRNVS). To solve this problem, we propose a novel framework using cross-guided optimization of the single-image super-resolution (SISR) and radiance fields. We perform multi-view image super-resolution (MVSR) on train-view images during the radiance fields optimization process. It derives the updated SR result by fusing the feature map obtained from SISR and voxel-based uncertainty fields generated by integrated errors of train-view images. By repeating the updates during radiance fields optimization, train-view images for radiance fields optimization have multi-view consistency and high-frequency details simultaneously, ultimately improving the performance of HRNVS. Experiments of HRNVS and MVSR on various benchmark datasets show that the proposed method significantly surpasses existing methods.

Motion deblurring from a blurred image is a challenging computer vision problem because frame-based cameras lose information during the blurring process. Several attempts have compensated for the loss of motion information by using event cameras, which are bio-inspired sensors with a high temporal resolution. Even though most studies have assumed that image and event data are pixel-wise aligned, this is only possible with low-quality active-pixel sensor (APS) images and synthetic datasets. In real scenarios, obtaining per-pixel aligned event-RGB data is technically challenging since event and frame cameras have different optical axes. For the application of the event camera, we propose the first Non-coaxial Event-guided Image Deblurring (NEID) approach that utilizes the camera setup composed of a standard frame-based camera with a non-coaxial single event camera. To consider the per-pixel alignment between the image and event without additional devices, we propose the first NEID network that spatially aligns events to images while refining the image features from temporally dense event features. For training and evaluation of our network, we also present the first large-scale dataset, consisting of RGB frames with non-aligned events aimed at a breakthrough in motion deblurring with an event camera. Extensive experiments on various datasets demonstrate that the proposed method achieves significantly better results than the prior works in terms of performance and speed, and it can be applied for practical uses of event cameras.

Recognizing objects from sparse and noisy events becomes extremely difficult when paired images and category labels do not exist. In this paper, we study label-free event-based object recognition where category labels and paired images are not available. To this end, we propose a joint formulation of object recognition and image reconstruction in a complementary manner. Our method first reconstructs images from events and performs object recognition through Contrastive Language-Image Pretraining (CLIP), enabling better recognition through a rich context of images. Since the category information is essential in reconstructing images, we propose category-guided attraction loss and category-agnostic repulsion loss to bridge the textual features of predicted categories and the visual features of reconstructed images using CLIP. Moreover, we introduce a reliable data sampling strategy and local-global reconstruction consistency to boost joint learning of two tasks. To enhance the accuracy of prediction and quality of reconstruction, we also propose a prototype-based approach using unpaired images. Extensive experiments demonstrate the superiority of our method and its extensibility for zero-shot object recognition. Our project code is available at https://github.com/Chohoonhee/Ev-LaFOR.

Point cloud completion is to restore complete 3D scenes and objects from incomplete observations or limited sensor data. Existing fully-supervised methods rely on paired datasets of incomplete and complete point clouds, which are labor-intensive to obtain. Unpaired methods have been proposed, but still require a set of complete point clouds as a reference. As a remedy, in this paper, we propose a novel point cloud completion framework without using any complete point cloud at all. Our main idea is to generate multiple incomplete point clouds of various poses and integrate them into a complete point cloud. We train our framework based on cycle consistency, to generate an incomplete point cloud such that 1) shares the same object as the input incomplete point cloud and 2) corresponds to an arbitrarily given pose. In addition, we devise a novel projection method conditioned by pose to gather visible features, from a volumetric feature extracted by an encoder. Extensive experiments demonstrate that the proposed method achieves comparable or better results than existing unpaired methods. Further, we show that our method also can be applied to real incomplete point clouds.

Our MATE is the first Test-Time-Training (TTT) method designed for 3D data, which makes deep networks trained for point cloud classification robust to distribution shifts occurring in test data. Like existing TTT methods from the 2D image domain, MATE also leverages test data for adaptation. Its test-time objective is that of a Masked Autoencoder: a large portion of each test point cloud is removed before it is fed to the network, tasked with reconstructing the full point cloud. Once the network is updated, it is used to classify the point cloud. We test MATE on several 3D object classification datasets and show that it significantly improves robustness of deep networks to several types of corruptions commonly occurring in 3D point clouds. We show that MATE is very efficient in terms of the fraction of points it needs for the adaptation. It can effectively adapt given as few as 5% of tokens of each test sample, making it extremely lightweight. Our experiments show that MATE also achieves competitive performance by adapting sparsely on the test data, which further reduces its computational overhead, making it ideal for real-time applications.

Reference-based video super-resolution (RefVSR) is a promising domain of super-resolution that recovers high-frequency textures of a video using reference video. The multiple cameras with different focal lengths in mobile devices aid recent works in RefVSR, which aim to super-resolve a low-resolution ultra-wide video by utilizing wide-angle videos. Previous works in RefVSR used all reference frames of a Ref video at each time step for the super-resolution of low-resolution videos. However, computation on higher-resolution images increases the runtime and memory consumption, hence hinders the practical application of RefVSR. To solve this problem, we propose an Efficient Reference-based Video Super-Resolution (ERVSR) that exploits a single reference frame to super-resolve whole low-resolution video frames. We introduce an attention-based feature align module and an aggregation upsampling module that attends LR features using the correlation between the reference and LR frames. The proposed ERVSR achieves 12× faster speed, 1/4 memory consumption than previous state-of-the-art RefVSR networks, and competitive performance on the RealMCVSR dataset while using a single reference image.

Event cameras asynchronously output the polarity values of pixel-level log intensity alterations. They are robust against motion blur and can be adopted in challenging light conditions. Owing to these advantages, event cameras have been employed in various vision tasks such as depth estimation, visual odometry, and object detection. In particular, event cameras are effective in stereo depth estimation to find correspondence points between two cameras under challenging illumination conditions and/or fast motion. However, because event cameras provide spatially sparse event stream data, it is difficult to obtain a dense disparity map. Although it is possible to estimate disparity from event data at the edge of a structure where intensity changes are likely to occur, estimating the disparity in a region where event occurs rarely is challenging. In this study, we propose a deep network that combines the features of an image with the features of an event to generate a dense disparity map. The proposed network uses images to obtain spatially dense features that are lacking in events. In addition, we propose a spatial multi-scale correlation between two fused feature maps for an accurate disparity map. To validate our method, we conducted experiments using synthetic and real-world datasets.

Learning to estimate object pose often requires ground-truth (GT) labels, such as CAD model and absolute-scale object pose, which is expensive and laborious to obtain in the real world. To tackle this problem, we propose an unsupervised domain adaptation (UDA) for category-level object pose estimation, called UDA-COPE. Inspired by recent multi-modal UDA techniques, the proposed method exploits a teacher-student self-supervised learning scheme to train a pose estimation network without using target domain pose labels. We also introduce a bidirectional filtering method between the predicted normalized object coordinate space (NOCS) map and observed point cloud, to not only make our teacher network more robust to the target domain but also to provide more reliable pseudo labels for the student network training. Extensive experimental results demonstrate the effectiveness of our proposed method both quantitatively and qualitatively. Notably, without leveraging target-domain GT labels, our proposed method achieved comparable or sometimes superior performance to existing methods that depend on the GT labels.

Neuromorphic cameras or event cameras mimic human vision by reporting changes in the intensity in a scene, instead of reporting the whole scene at once in a form of an image frame as performed by conventional cameras. Events are streamed data that are often dense when either the scene changes or the camera moves rapidly. The rapid movement causes the events to be overridden or missed when creating a tensor for the machine to learn on. To alleviate the event missing or overriding issue, we propose to learn to concentrate on the dense events to produce a compact event representation with high details for depth estimation. Specifically, we learn a model with events from both past and future but infer only with past data with the predicted future. We initially estimate depth in an event-only setting but also propose to further incorporate images and events by a hier-archical event and intensity combination network for better depth estimation. By experiments in challenging real-world scenarios, we validate that our method outperforms prior arts even with low computational cost. Code is available at: https://github.com/yonseivnl/se-cff.

Test-time adaptation approaches have recently emerged as a practical solution for handling domain shift without access to the source domain data. In this paper, we propose and explore a new multi-modal extension of test-time adaptation for 3D semantic segmentation. We find that, directly applying existing methods usually results in performance instability at test time, because multi-modal input is not considered jointly. To design a framework that can take full advantage of multi-modality, where each modality provides regularized self-supervisory signals to other modalities, we propose two complementary modules within and across the modalities. First, Intra-modal Pseudo-label Generation (Intra-PG) is introduced to obtain reliable pseudo labels within each modality by aggregating information from two models that are both pre-trained on source data but updated with target data at different paces. Second, Inter-modal Pseudo-label Refinement (Inter-PR) adaptively selects more reliable pseudo labels from different modalities based on a proposed consistency scheme. Experiments demonstrate that our regularized pseudo labels produce stable self-learning signals in numerous multi-modal test-time adaptation scenarios for 3D semantic segmentation. Visit our project website at https://www.nec-labs.com/~mas/MM-TTA

Visual degradations caused by motion blur, raindrop, rain, snow, illumination, and fog deteriorate image quality and, subsequently, the performance of perception algorithms deployed in outdoor conditions. While degradation-specific image restoration techniques have been extensively studied, such algorithms are domain sensitive and fail in real scenarios where multiple degradations exist simultaneously. This makes a case for blind image restoration and reconstruction algorithms as practically relevant. However, the absence of a dataset diverse enough to encapsulate all variations hinders development for such an algorithm. In this paper, we utilize a synthetic degradation model that recursively applies sets of random degradations to generate naturalistic degradation images of varying complexity, which are used as input. Furthermore, as the degradation intensity can vary across an image, the spatially invariant convolutional filter cannot be applied for all degradations. Hence to enable spatial variance during image restoration and reconstruction, we design a transformer-based architecture to benefit from the long-range dependencies. In addition, to reduce the computational cost of transformers, we propose a multi-branch structure coupled with modifications such as a complimentary feature selection mechanism and the replacement of a feed-forward network with lightweight multiscale convolutions. Finally, to improve restoration and reconstruction, we integrate an auxiliary decoder branch to predict the degradation mask to ensure the underlying network can localize the degradation information. From empirical analysis on 10 datasets covering rain drop removal, deraining, dehazing, image enhancement, and deblurring, we demonstrate the efficacy of the proposed approach while obtaining SoTA performance.

The$360^{\circ}$imaging has recently gained much attention; however, its angular resolution is relatively lower than that of a narrow field-of-view (FOV) perspective image as it is captured using a fisheye lens with the same sensor size. Therefore, it is beneficial to super-resolve a$360^{\circ}$image. Several attempts have been made, but mostly considered equirectangular projection (ERP) as one of the ways for$360^{\circ}$image representation despite the latitude-dependent distortions. In that case, as the output high-resolution (HR) image is always in the same ERP format as the low-resolution (LR) input, additional information loss may occur when transforming the HR image to other projection types. In this paper, we propose SphereSR, a novel framework to generate a continuous spherical image representation from an LR$360^{\circ}$image, with the goal of predicting the RGB values at given spherical coordinates for super-resolution with an arbitrary$360^{\circ}$image projection. Specifically, first we propose a feature extraction module that represents the spherical data based on an icosahedron and that efficiently extracts features on the spherical surface. We then propose a spherical local implicit image function (SLIIF) to predict RGB values at the spherical coordinates. As such, SphereSR flexibly reconstructs an HR image given an arbitrary projection type. Experiments on various benchmark datasets show that the proposed method significantly surpasses existing methods in terms of performance.

Semantic segmentation provides scene understanding capability by performing pixel-wise classification of objects within an image. However, the sensitivity of such algorithms towards domain changes requires fine-tuning using an annotated dataset for each novel domain, which is expensive to construct and inefficient. We highlight that irrespective of the training dataset, structural properties of scenes remain the same hence domain sensitivity arises from training methodology. Thus, in this paper, we propose a domain alignment approach wherein multiple synthetic source domains are used to train an underlying segmentation network such that it performs consistently in unknown real target domains. Towards this end, we propose a pixel-wise supervised contrastive learning framework that enforces constraints in latent space resulting in features belonging to the same class being clustered closely and away from different classes. This approach allows for better capturing of global and local semantics while providing domain invariant properties. Our approach can be easily incorporated into prior semantic segmentation approaches without the significant computational overhead. We empirically demonstrate the efficacy of the proposed approach on GTAV → Cityscapes, GTAV+Synthia → Cityscapes, and GTAV+Synthia+Synscapes → Cityscapes scenarios and report state-of-the-art (SoTA) performance without requiring access to images from the target domain.

Increasing usage of traffic cameras provides an opportunity to utilize them for smart city applications. However, the efficacy of such systems is determined by their ability to detect and track objects of interest from diverse viewpoints accurately. This is challenging due to the diverse viewpoints, elevations, and distinct properties of camera sensors. Thus, to ensure robust performance, the training dataset should cover many variations, including viewpoints, illumination changes, and diverse weather conditions. However, constructing such a dataset is expensive in terms of data collection and annotation. This paper proposes an unsupervised domain adaptation approach wherein a synthetic dataset is generated using a simulator and subsequently used to ensure performance consistency of multi-object-tracking (MOT) algorithms across a diverse range of manually annotated natural scenes. Towards this end, we emphasize achieving domain invariant object detection by combining image stylization and class-balancing augmentation. Furthermore, we extend the robust detection algorithm to track detected objects across a large time scale using feature embeddings generated by the detector. Based on qualitative and quantitative results, we demonstrate the viability of such a system that is invariant to illumination, weather, viewpoint, and scene changes while providing a baseline for future research. Codebase and datasets would be made available at https://github.com/pranjay-dev/IS2R.

The aim of weakly supervised semantic segmentation (WSSS) is to learn semantic segmentation without using dense annotations. WSSS has been intensively studied for 2D images and 3D point clouds. However, the existing WSSS studies have focused on a single domain, i.e. 2D or 3D, even when multi-domain data is available. In this paper, we propose a novel joint 2D-3D WSSS framework taking advantage of WSSS in different domains, using classification labels only. Via projection, we leverage the 2D class activation map as self-supervision to enhance the 3D semantic perception. Conversely, we exploit the similarity matrix of point cloud features for training the image classifier to achieve more precise 2D segmentation. In both directions, we devise a confidence-based scoring method to reduce the effect of inaccurate self-supervision. With extensive quantitative and qualitative experiments, we verify that the proposed joint WSSS framework effectively transfers the benefit of each domain to the other domain, and the resulting semantic segmentation performance is remarkably improved in both 2D and 3D domains. On the ScanNetV2 benchmark, our framework significantly outperforms the prior WSSS approaches, suggesting a new research direction for WSSS.

An event camera reports per-pixel intensity differences as an asynchronous stream of events with low latency, high dynamic range (HDR), and low power consumption. This stream of sparse/dense events limits the direct use of well-known computer vision applications for event cameras. Further applications of event streams to vision tasks that are sensitive to image quality issues, such as spatial resolution and blur, e.g., object detection, would benefit from a higher resolution of image reconstruction. Moreover, despite the recent advances in spatial resolution in event camera hardware, the majority of commercially available event cameras still have relatively low spatial resolutions when compared to conventional cameras. We propose an end-to-end recurrent network to reconstruct high-resolution, HDR, and temporally consistent grayscale or color frames directly from the event stream, and extend it to generate temporally consistent videos. We evaluate our algorithm on real-world and simulated sequences and verify that it reconstructs fine details of the scene, outperforming previous methods in quantitative quality measures. We further investigate how to (1) incorporate active pixel sensor frames (produced by an event camera) and events together in a complementary setting and (2) reconstruct images iteratively to create an even higher quality and resolution in the images.

Omni-directional images are becoming more prevalent for understanding the scene of all directions around a camera, as they provide a much wider field-of-view (FoV) compared to conventional images. In this work, we present a novel approach to represent omni-directional images and suggest how to apply CNNs on the proposed image representation. The proposed image representation method utilizes a spherical polyhedron to reduce distortion introduced inevitably when sampling pixels on a non-Euclidean spherical surface around the camera center. To apply convolution operation on our representation of images, we stack the neighboring pixels on top of each pixel and multiply with trainable parameters. This approach enables us to apply the same CNN architectures used in conventional euclidean 2D images on our proposed method in a straightforward manner. Compared to the previous work, we additionally compare different designs of kernels that can be applied to our proposed method. We also show that our method outperforms in monocular depth estimation task compared to other state-of-the-art representation methods of omni-directional images. In addition, we propose a novel method to fit bounding ellipses of arbitrary orientation using object detection networks and apply it to an omni-directional real-world human detection dataset.

Event cameras sense brightness changes in each pixel and yield asynchronous event streams instead of producing intensity images. They have distinct advantages over conventional cameras, such as a high dynamic range (HDR) and no motion blur. To take advantage of event cameras with existing image-based algorithms, a few methods have been proposed to reconstruct images from event streams. However, the output images have a low resolution (LR) and are unrealistic. Low-quality outputs stem from broader applications of event cameras, where high-quality and high-resolution (HR) images are needed. In this work, we consider the problem of reconstructing and super-resolving images from LR events when no ground truth (GT) HR images and degradation models are available. We propose a novel end-to-end joint framework for single image reconstruction and super-resolution from LR event data. Our method is primarily unsupervised to handle the absence of real inputs from GT and deploys adversarial learning. To train our framework, we constructed an open dataset, including simulated events and real-world images. The use of the dataset boosts the network performance, and the network architectures and various loss functions in each phase help improve the quality of the resulting image. Various experiments showed that our method surpasses the state-of-the-art LR image reconstruction methods for real-world and synthetic datasets. The experiments for super-resolution (SR) image reconstruction also substantiate the effectiveness of the proposed method. We further extended our method to more challenging problems of HDR, sharp image reconstruction, and color events. In addition, we demonstrate that the reconstruction and super-resolution results serve as intermediate representations of events for high-level tasks, such as semantic segmentation, object recognition, and detection. We further examined how events affect the outputs of the three phases and analyze our method's efficacy through an ablation study.

Deep neural models, in recent years, have been successful in almost every field, even solving the most complex problem statements. However, these models are huge in size with millions (and even billions) of parameters, demanding heavy computation power and failing to be deployed on edge devices. Besides, the performance boost is highly dependent on redundant labeled data. To achieve faster speeds and to handle the problems caused by the lack of labeled data, knowledge distillation (KD) has been proposed to transfer information learned from one model to another. KD is often characterized by the so-called 'Student-Teacher' (S-T) learning framework and has been broadly applied in model compression and knowledge transfer. This paper is about KD and S-T learning, which are being actively studied in recent years. First, we aim to provide explanations of what KD is and how/why it works. Then, we provide a comprehensive survey on the recent progress of KD methods together with S-T frameworks typically used for vision tasks. In general, we investigate some fundamental questions that have been driving this research area and thoroughly generalize the research progress and technical details. Additionally, we systematically analyze the research status of KD in vision applications. Finally, we discuss the potentials and open challenges of existing methods and prospect the future directions of KD and S-T learning.

High dynamic range (HDR) imaging is a technique that allows an extensive dynamic range of exposures, which is important in image processing, computer graphics, and computer vision. In recent years, there has been a significant advancement in HDR imaging using deep learning (DL). This study conducts a comprehensive and insightful survey and analysis of recent developments in deep HDR imaging methodologies. We hierarchically and structurally group existing deep HDR imaging methods into five categories based on (1) number/domain of input exposures, (2) number of learning tasks, (3) novel sensor data, (4) novel learning strategies, and (5) applications. Importantly, we provide a constructive discussion on each category regarding its potential and challenges. Moreover, we review some crucial aspects of deep HDR imaging, such as datasets and evaluation metrics. Finally, we highlight some open problems and point out future research directions.

Presence of haze in images obscures underlying information, which is undesirable in applications requiring accurate environment information. To recover such an image, a dehazing algorithm should localize and recover affected regions while ensuring consistency between recovered and its neighboring regions. However owing to fixed receptive field of convolutional kernels and non uniform haze distribution, assuring consistency between regions is difficult. In this paper, we utilize an encoder-decoder based network architecture to perform the task of dehazing and integrate an spatially aware channel attention mechanism to enhance features of interest beyond the receptive field of traditional conventional kernels. To ensure performance consistency across diverse range of haze densities, we utilize greedy localized data augmentation mechanism. Synthetic datasets are typically used to ensure a large amount of paired training samples, however the methodology to generate such samples introduces a gap between them and real images while accounting for only uniform haze distribution and overlooking more realistic scenario of non-uniform haze distribution resulting in inferior dehazing performance when evaluated on real datasets. Despite this, the abundance of paired samples within synthetic datasets cannot be ignored. Thus to ensure performance consistency across diverse datasets, we train the proposed network within an adversarial prior-guided framework that relies on a generated image along with its low and high frequency components to determine if properties of dehazed images matches those of ground truth. We preform extensive experiments to validate the dehazing and domain invariance performance of proposed framework across diverse domains and report state-of-the-art (SoTA) results. The source code with pretrained models will be available at https://github.com/PS06/DIDH.

Event cameras sense per-pixel intensity changes and produce asynchronous event streams with high dynamic range and less motion blur, showing advantages over the conventional cameras. A hurdle of training event-based models is the lack of large qualitative labeled data. Prior works learning end-tasks mostly rely on labeled or pseudo-labeled datasets obtained from the active pixel sensor (APS) frames; however, such datasets’ quality is far from rivaling those based on the canonical images. In this paper, we propose a novel approach, called EvDistill, to learn a student network on the unlabeled and unpaired event data (target modality) via knowledge distillation (KD) from a teacher network trained with large-scale, labeled image data (source modality). To enable KD across the unpaired modalities, we first propose a bidirectional modality reconstruction (BMR) module to bridge both modalities and simultaneously exploit them to distill knowledge via the crafted pairs, causing no extra computation in the inference. The BMR is improved by the end-tasks and KD losses in an end-to-end manner. Second, we leverage the structural similarities of both modalities and adapt the knowledge by matching their distributions. Moreover, as most prior feature KD methods are uni-modality and less applicable to our problem, we propose an affinity graph KD loss to boost the distillation. Our extensive experiments on semantic segmentation and object recognition demonstrate that EvDistill achieves significantly better results than the prior works and KD with only events and APS frames.

Event cameras are novel sensors that perceive the perpixel intensity changes and output asynchronous event streams with high dynamic range and less motion blur. It has been shown that events alone can be used for end-task learning, e.g., semantic segmentation, based on encoder-decoder-like networks. However, as events are sparse and mostly reflect edge information, it is difficult to recover original details merely relying on the decoder. Moreover, most methods resort to the pixel-wise loss alone for supervision, which might be insufficient to fully exploit the visual details from sparse events, thus leading to less optimal performance. In this paper, we propose a simple yet flexible two-stream framework named Dual Transfer Learning (DTL) to effectively enhance the performance on the end-tasks without adding extra inference cost. The proposed approach consists of three parts: event to end-task learning (EEL) branch, event to image translation (EIT) branch, and transfer learning (TL) module that simultaneously explores the feature-level affinity information and pixel-level knowledge from the EIT branch to improve the EEL branch. This simple yet novel method leads to strong representation learning from events and is evidenced by the significant performance boost on the end-tasks such as semantic segmentation and depth estimation.

Event cameras can report scene movements as an asynchronous stream of data called the events. Unlike traditional cameras, event cameras have very low latency (microseconds vs milliseconds) very high dynamic range (140 dB vs 60 dB), and low power consumption, as they report changes of a scene and not a complete frame. As they re- port per pixel feature-like events and not the whole intensity frame they are immune to motion blur. However, event cameras require movement between the scene and camera to fire events, i.e., they have no output when the scene is relatively static. Traditional cameras, however, report the whole frame of pixels at once in fixed intervals but have lower dynamic range and are prone to motion blur in case of rapid movements. We get the best from both worlds and use events and intensity images together in our complementary design and estimate dense disparity from this combination. The proposed end-to-end design combines events and images in a sequential manner and correlates them to estimate dense depth values. Our various experimental settings in real-world and simulated scenarios exploit the superiority of our method in predicting accurate depth values with fine details. We further extend our method to extreme cases of missing the left or right event or stereo pair and also investigate stereo depth estimation with inconsistent dynamic ranges or event thresholds on the left and right pairs.

Weakly supervised semantic segmentation (WSSS) using image-level classification labels usually utilizes the Class Activation Maps (CAMs) to localize objects of interest in images. While pointing out that CAMs only highlight the most discriminative regions of the classes of interest, adversarial erasing (AE) methods have been proposed to further explore the less discriminative regions. In this paper, we review the potential of the pre-trained classifier which is trained on the raw images. We experimentally verify that the ordinary classifier1already has the capability to activate the less discriminative regions if the most discriminative regions are erased to some extent. Based on that, we propose a class-specific AE-based framework that fully exploits the potential of an ordinary classifier. Our framework (1) adopts the ordinary classifier to notify the regions to be erased and (2) generates a class-specific mask for erasing by randomly sampling a single specific class to be erased (target class) among the existing classes on the image for obtaining more precise CAMs. Specifically, with the guidance of the ordinary classifier, the proposed CAMs Generation Network (CGNet) is enforced to generate a CAM of the target class while constraining the CAM not to intrude the object regions of the other classes. Along with the pseudo-labels refined from our CAMs, we achieve the state-of-the-art WSSS performance on both PASCAL VOC 2012 and MS-COCO dataset only with image-level supervision. The code is available at https://github.com/KAIST-vilab/OC-CSE.

Existing vanishing point (VP) estimation methods rely on pre-extracted image lines and/or prior knowledge of the number of VPs. However, in practice, this information may be insufficient or unavailable. To solve this problem, we propose a network that treats a perspective image as input and predicts a spherical probability map of VP. Based on this map, we can detect all the VPs. Our method is reliable thanks to four technical novelties. First, we leverage the icosahedral spherical representation to express our probability map. This representation provides uniform pixel distribution, and thus facilitates estimating arbitrary positions of VPs. Second, we design a loss function that enforces the antipodal symmetry and sparsity of our spherical probability map to prevent over-fitting. Third, we generate the ground truth probability map that reasonably expresses the locations and uncertainties of VPs. This map unnecessarily peaks at noisy annotated VPs, and also exhibits various anisotropic dispersions. Fourth, given a predicted probability map, we detect VPs by fitting a Bingham mixture model. This strategy can robustly handle close VPs and provide the confidence level of VP useful for practical applications. Experiments showed that our method achieves the best compromise between generality, accuracy, and efficiency, compared with state-of-the-art approaches.

Vehicle Re-identification (Re-ID) aims to retrieve all instances of query vehicle images present in an image pool. However viewpoint, illumination, and occlusion variations along with subtle differences between two unique images pose a significant challenge towards achieving an effective system. In this paper, we emphasize upon enhancing the performance of visual feature based ReID system by improving feature embedding quality and propose (1) an attention-guided hierarchical feature extractor (HFE) that leverages the structure of a backbone CNN to extract coarse and fine-grained features and (2) to train the proposed network within a hard negative adversarial framework that generates samples exhibiting extreme variations, encouraging the network to extract important distinguishing features across varying scales. To demonstrate the effectiveness of the proposed framework we use VERI-Wild, VRIC and Veri-776 datasets that exhibit extreme intra-class and minute inter-class differences and achieve state-of-the-art (SoTA) performance. Codes related to this paper are publicly available at https://github.com/PS06/VReID.

Recent advances in deep neural networks (DNNs) have facilitated high-end applications, including holistic scene understanding (HSU), in which many tasks run in parallel with the same visual input. Following this trend, various methods have been proposed to use DNNs to perform multiple vision tasks. However, these methods are task-specific and less effective when considering multiple HSU tasks. End-to-end demonstrations of adversarial examples, which generate one-to-many heterogeneous adversarial examples in parallel from the same input, are scarce. Additionally, one-to-many mapping of adversarial examples for HSU usually requires joint representation learning and flexible constraints on magnitude, which can render the prevalent attack methods ineffective. In this paper, we propose PSAT-GAN, an end-to-end framework that follows the pipeline of HSU. It is based on a mixture of generative models and an adversarial classifier that employs partial weight sharing to learn a one-to-many mapping of adversarial examples in parallel, each of which is effective for its corresponding task in HSU attacks. PSAT-GAN is further enhanced by applying novel adversarial and soft-constraint losses to generate effective perturbations and avoid studying transferability. Experimental results indicate that our method is efficient in generating both universal and image-dependent adversarial examples to fool HSU tasks under either targeted or non-targeted settings.

Event cameras sense intensity changes and have many advantages over conventional cameras. To take advantage of event cameras, some methods have been proposed to reconstruct intensity images from event streams. However, the outputs are still in low resolution (LR), noisy, and unrealistic. The low-quality outputs stem broader applications of event cameras, where high spatial resolution (HR) is needed as well as high temporal resolution, dynamic range, and no motion blur. We consider the problem of reconstructing and super-resolving intensity images from pure events, when no ground truth (GT) HR images and down-sampling kernels are available. To tackle the challenges, we propose a novel end-to-end pipeline that reconstructs LR images from event streams, enhances the image qualities and upsamples the enhanced images, called EventSR. For the absence of real GT images, our method is primarily unsupervised, deploying adversarial learning. To train EventSR, we create an open dataset including both real-world and simulated scenes. The use of both datasets boosts up the network performance, and the network architectures and various loss functions in each phase help improve the image qualities. The whole pipeline is trained in three phases. While each phase is mainly for one of the three tasks, the networks in earlier phases are fine-tuned by respective loss functions in an end-to-end manner. Experimental results show that EventSR generates high-quality SR images from events for both simulated and real-world data.

An event camera detects per-pixel intensity difference and produces asynchronous event stream with low latency, high dynamic range, and low power consumption. As a trade-off, the event camera has low spatial resolution. We propose an end-to-end network to reconstruct high resolution, high dynamic range (HDR) images directly from the event stream. We evaluate our algorithm on both simulated and real-world sequences and verify that it captures fine details of a scene and outperforms the combination of the state-of-the-art event to image algorithms with the state-of-the-art super resolution schemes in many quantitative measures by large margins. We further extend our method by using the active sensor pixel (APS) frames or reconstructing images iteratively.

Anchor boxes act as potential object localization candidates allow single-stage detectors to achieve real-time performance, at the cost of localization accuracy when compared to state-of-the-art two-stage detectors. Therefore, correct selection of the scale and aspect ratio associated with an anchor box is crucial for detector performance. In this work, we propose a novel architecture called DANet for improving the localization performance of single-stage object detectors, while maintaining real-time inference. The proposed network achieves this by predicting (1) the combination of aspect ratio and scale per feature map based on object density and (2) localization confidence per anchor box. We evaluate the proposed network using the benchmark dataset. On the MS COCO dataset, DANet achieves 30.9% AP at 51.8 fps using ResNet-18 and 45.3% AP at 7.4 fps using ResNeXt-101. The code and models will be available at https://github.com/PS06/AnchorNet.

We propose an unsupervised approach to predict depth maps from images captured by a wireless endoscopic capsule. Recent advances in deep learning have shown that accurate depth maps can be predicted from a single image, where the deep network is trained via unsupervised or self-supervised learning by using monocular video sequences or stereo image pairs. However, directly applying these techniques to endoscopic images does not yield satisfactory results owing to the inherent difficulties of the wireless capsule imaging such as dim lighting and low-resolution of images, which are different from normal imaging conditions. For that reason, we exploit the environmental characteristics of endoscopic images - there is no external light source except ones attached to the capsule. Based on this condition, we propose the direct attenuation model-based depth map prediction scheme to guide depth prediction and to add meaningful cues to the loss function. We experimentally verify the proposed method with various endoscopic images.

Omni-directional cameras have many advantages over conventional cameras in that they have a much wider field-of-view (FOV). Accordingly, several approaches have been proposed recently to apply convolutional neural networks (CNNs) to omni-directional images for various visual tasks. However, most of them use image representations defined in the Euclidean space after transforming the omni-directional views originally formed in the non-Euclidean space. This transformation leads to shape distortion due to nonuniform spatial resolving power and the loss of continuity. These effects make existing convolution kernels experience difficulties in extracting meaningful information. This paper presents a novel method to resolve such problems of applying CNNs to omni-directional images. The proposed method utilizes a spherical polyhedron to represent omni-directional views. This method minimizes the variance of the spatial resolving power on the sphere surface, and includes new convolution and pooling methods for the proposed representation. The proposed method can also be adopted by any existing CNN-based methods. The feasibility of the proposed method is demonstrated through classification, detection, and semantic segmentation tasks with synthetic and real datasets.

Event cameras have a lot of advantages over traditional cameras, such as low latency, high temporal resolution, and high dynamic range. However, since the outputs of event cameras are the sequences of asynchronous events over time rather than actual intensity images, existing algorithms could not be directly applied. Therefore, it is demanding to generate intensity images from events for other tasks. In this paper, we unlock the potential of event camera-based conditional generative adversarial networks to create images/videos from an adjustable portion of the event data stream. The stacks of space-time coordinates of events are used as inputs and the network is trained to reproduce images based on the spatio-temporal intensity changes. The usefulness of event cameras to generate high dynamic range (HDR) images even in extreme illumination conditions and also non blurred images under rapid motion is also shown. In addition, the possibility of generating very high frame rate videos is demonstrated, theoretically up to 1 million frames per second(FPS) since the temporal resolution of event cameras is about 1 microsecond. Proposed methods are evaluated by comparing the results with the intensity images captured on the same pixel grid-line of events using online available real datasets and synthetic datasets produced by the event camera simulator.

We present a robust approach for computing disparity maps with a supervised learning-based confidence prediction. This approach takes into consideration following features. First, we analyze the characteristics of various confidence measures in the random forest framework to select effective confidence measures depending on the characteristics of the training data and matching strategies, such as similarity measures and parameters. We then train a random forest using the selected confidence measures to improve the efficiency of confidence prediction and to build a better prediction model. Second, we present a confidence-based matching cost modulation scheme, based on predicted confidence values, to improve the robustness and accuracy of the (semi-) global stereo matching algorithms. Finally, we apply the proposed modulation scheme to popularly used algorithms to make them robust against unexpected difficulties that could occur in an uncontrolled environment using challenging outdoor datasets. The proposed confidence measure selection and cost modulation schemes are experimentally verified from various perspectives using the KITTI and Middlebury datasets.

Recently developed object detectors rely on automatically generated object proposals, instead of using a dense sliding window search scheme; generating good object proposals has therefore become crucial for improving the computational cost and accuracy of object detection performance. In particular, the shape and location errors of object proposals can be directly propagated to object detection unless some additional processes are adopted to refine the shape and location of bounding boxes. In this study, the authors demonstrate an object proposal refinement algorithm that improves the localisation accuracy and refines the shape of object proposals by searching a boundary‐aligned minimum bounding box. They assume that an object consists of several image regions, and that the optimal object proposal is well aligned with image region boundaries. Based on this assumption, they design novel boundary‐region alignment measures and then propose a greedy refinement method based on the proposed measures. Experiments on the PASCAL VOC 2007 dataset show that the proposed method produces highly well‐localised object proposals and truly improves the quality of object proposals.

Online multi-object tracking aims at estimating the tracks of multiple objects instantly with each incoming frame and the information provided up to the moment. It still remains a difficult problem in complex scenes, because of the large ambiguity in associating multiple objects in consecutive frames and the low discriminability between objects appearances. In this paper, we propose a robust online multi-object tracking method that can handle these difficulties effectively. We first define the tracklet confidence using the detectability and continuity of a tracklet, and decompose a multi-object tracking problem into small subproblems based on the tracklet confidence. We then solve the online multi-object tracking problem by associating tracklets and detections in different ways according to their confidence values. Based on this strategy, tracklets sequentially grow with online-provided detections, and fragmented tracklets are linked up with others without any iterative and expensive association steps. For more reliable association between tracklets and detections, we also propose a deep appearance learning method to learn a discriminative appearance model from large training datasets, since the conventional appearance learning methods do not provide rich representation that can distinguish multiple objects with large appearance variations. In addition, we combine online transfer learning for improving appearance discriminability by adapting the pre-trained deep model during online tracking. Experiments with challenging public datasets show distinct performance improvement over other state-of-the-arts batch and online tracking methods, and prove the effect and usefulness of the proposed methods for online multi-object tracking.

Person re-identification is the problem of recognizing people across different images or videos with non-overlapping views. Although a significant progress has been made in person re-identification over the last decade, it remains a challenging task because the appearances of people can seem extremely different across diverse camera viewpoints and person poses. In this paper, we propose a novel framework for person re-identification by analyzing camera viewpoints and person poses called pose-aware multi-shot matching. It robustly estimates individual poses and efficiently performs multi-shot matching based on the pose information. The experimental results obtained by using public person re-identification data sets show that the proposed methods outperform the current state-of-the-art methods, and are promising for accomplishing person re-identification under diverse viewpoints and pose variances.

This paper addresses the multi-attributed graph matching problem, which considers multiple attributes jointly while preserving the characteristics of each attribute for graph matching. Since most of conventional graph matching algorithms integrate multiple attributes to construct a single unified attribute in an oversimplified manner, the information from multiple attributes is often not completely utilized. In order to solve this problem, we propose a novel multi-layer graph structure that can preserve the characteristics of each attribute in separated layers, and also propose a multi-attributed graph matching algorithm based on random walk centrality with the proposed multi-layer graph structure. We compare the proposed algorithm with other state-of-the-art graph matching algorithms based on a single-layer structure using synthetic and real data sets and demonstrate the superior performance of the proposed multi-layer graph structure and the multi-attributed graph matching algorithm.

Road surface profile (RSP) estimation is an important task to find the imperfections of a road surface, thereby improving ride quality. The RSP estimation has been recently studied using stereo vision owing to its affordable price. However, the existing methods provide noisy and temporally unstable results for real-world driving scenes because of noisy range measurements, noisy pitch angle estimates between the camera and the road surface, and interference from obstacles. This paper proposes a novel method for temporally consistent and robust RSP estimation to overcome these problems. The proposed method consists of three steps: free space estimation, digital elevation map (DEM) estimation, and RSP estimation. We first estimate a drivable area, i.e., free space. For robust and fast free space estimation, we propose an optimization-based non-parametric road surface modeling method and an integral disparity histogram-based free space estimation method. Then, the DEM of the road surface is estimated using range measurements on the free space to avoid obstacle interference. The DEM is temporally updated every frame using the moving average filter and the DEM reference grid update scheme. Owing to these strategies, the proposed method reduces the elevation estimation noise and the pitch angle error, and therefore provides the temporally consistent RSP. We demonstrate the superiority of the proposed method experimentally using stereo image sequences captured in real-world driving scenes.

To watch 360° videos on normal 2D displays, we need to project the selected part of the 360° image onto the 2D display plane. In this paper, we propose a fully-automated framework for generating content-aware 2D normal-view perspective videos from 360° videos. Especially, we focus on the projection step preserving important image contents and reducing image distortion. Basically, our projection method is based on Pannini projection model. At first, the salient contents such as linear structures and salient regions in the image are preserved by optimizing the single Panini projection model. Then, the multiple Panini projection models at salient regions are interpolated to suppress image distortion globally. Finally, the temporal consistency for image projection is enforced for producing temporally stable normal-view videos. Our proposed projection method does not require any user-interaction and is much faster than previous content-preserving methods. It can be applied to not only images but also videos taking the temporal consistency of projection into account. Experiments on various 360° videos show the superiority of the proposed projection method quantitatively and qualitatively.

In this paper, we propose a novel method to jointly solve scene layout estimation and global registration problems for accurate indoor 3D reconstruction. Given a sequence of range data, we first build a set of scene fragments using KinectFusion and register them through pose graph optimization. Afterwards, we alternate between layout estimation and layout-based global registration processes in iterative fashion to complement each other. We extract the scene layout through hierarchical agglomerative clustering and energy-based multi-model fitting in consideration of noisy measurements. Having the estimated scene layout in one hand, we register all the range data through the global iterative closest point algorithm where the positions of 3D points that belong to the layout such as walls and a ceiling are constrained to be close to the layout. We experimentally verify the proposed method with the publicly available synthetic and real-world datasets in both quantitative and qualitative ways.

For 3D object detection and pose estimation, it is crucial to extract distinctive and representative features of the objects and describe them efficiently. Therefore, a large number of 3D feature descriptors has been developed. Among these, Point Feature Histogram RGB (PFHRGB) has been evaluated as showing the best performance for 3D object and category recognition. However, this descriptor is vulnerable to point density variation and produces many false correspondences accordingly. In this paper, we tackle this problem and propose an algorithm to find the correct correspondences under the point density variation. Experimental results show that the proposed method is promising for 3D object detection and pose estimation under the point density variation.

The detection of dynamic points on a moving platform is an important task to avoid a potential collision. However, it is difficult to detect dynamic points using only two frames, especially when various input data such as ego-motion, disparity map, and optical flow are noisy for computing the motion of points. In this paper, we propose a supervised learning-based approach to detect dynamic points in consideration of noisy input data. First of all, to consider depth ambiguity that proportionally increases according to the distance to the ego-vehicle, we divide the XZ-plane (bird-eye view) into several subregions. Then, we train a random forest for each subregion by constructing motion vectors computed based on two motion metrics. Here, in order to reduce errors of the input data, the motion vectors are filtered based on a pairwise planarity check and then filtered motion vectors are used for training. In the experiments, the proposed method is verified by comparing the detection performance with that of previous approaches on the KITTI dataset.

Person re-identification is the problem of recognizing people across images or videos from non-overlapping views. Although there has been much progress in person re-identification for the last decade, it still remains a challenging task because of severe appearance changes of a person due to diverse camera viewpoints and person poses. In this paper, we propose a novel framework for person reidentification by analyzing camera viewpoints and person poses, so-called Pose-aware Multi-shot Matching (PaMM), which robustly estimates target poses and efficiently conducts multi-shot matching based on the target pose information. Experimental results using public person reidentification datasets show that the proposed methods are promising for person re-identification under diverse viewpoints and pose variances.

Multi-object tracking (MOT) becomes more challenging when objects of interest have similar appearances. In that case, the motion cues are particularly useful for discriminating multiple objects. However, for online 2D MOT in scenes acquired from moving cameras, observable motion cues are complicated by global camera movements and thus not always smooth or predictable. To deal with such unexpected camera motion for online 2D MOT, a structural motion constraint between objects has been utilized thanks to its robustness to camera motion. In this paper, we propose a new data association method that effectively exploits structural motion constraints in the presence of large camera motion. In addition, to further improve the robustness of data association against mis-detections and false positives, a novel event aggregation approach is developed to integrate structural constraints in assignment costs for online MOT. Experimental results on a large number of datasets demonstrate the effectiveness of the proposed algorithm for online 2D MOT.

When tracking multiple objects in an image sequence, various difficulties such as occlusion, mis-detection, false detection, and abrupt camera motion often occur together. Nevertheless, previous methods on multi-object tracking generally focus on only one or two of them. For that reason, the previous methods could not handle various problematic situations, where multiple difficulties occur simultaneously. To overcome this limitation, we propose a unified framework that can handle such difficulties concurrently, where we effectively combine the confidence-based two-step data association and relative motion network with correlation filtering. We show that the proposed unified framework yields noticeable performance enhancement under various difficulties.

A robust algorithm is proposed for tracking a target object in dynamic conditions including motion blurs, illumination changes, pose variations, and occlusions. To cope with these challenging factors, multiple trackers based on different feature representations are integrated within a probabilistic framework. Each view of the proposed multiview (multi-channel) feature learning algorithm is concerned with one particular feature representation of a target object from which a tracker is developed with different levels of reliability. With the multiple trackers, the proposed algorithm exploits tracker interaction and selection for robust tracking performance. In the tracker interaction, a transition probability matrix is used to estimate dependencies between trackers. Multiple trackers communicate with each other by sharing information of sample distributions. The tracker selection process determines the most reliable tracker with the highest probability. To account for object appearance changes, the transition probability matrix and tracker probability are updated in a recursive Bayesian framework by reflecting the tracker reliability measured by a robust tracker likelihood function that learns to account for both transient and stable appearance changes. Experimental results on benchmark datasets demonstrate that the proposed interacting multiview algorithm performs robustly and favorably against state-of-the-art methods in terms of several quantitative metrics.

A widely-used approach for estimating camera orientation is to use points at infinity, i.e., vanishing points (VPs). By enforcing the orthogonal constraint between the VPs, called the Manhattan world constraint, a drift-free camera orientation estimation can be achieved. However, in practical applications this approach suffers from many spurious parallel line segments or does not perform in non-Manhattan world scenes. To overcome these limitations, we propose a novel method that jointly estimates the VPs and camera orientation based on sequential Bayesian filtering. The proposed method does not require the Manhattan world assumption, and can perform a highly accurate estimation of camera orientation in real time. In addition, in order to enhance the robustness of the joint estimation, we propose a feature management technique that removes false positives of line clusters and classifies newly detected lines. We demonstrate the superiority of the proposed method through an extensive evaluation using synthetic and real datasets and comparison with other state-of-the-art methods.

We propose a new approach to associate supervised learning-based confidence prediction with the stereo matching problem. First of all, we analyze the characteristics of various confidence measures in the regression forest framework to select effective confidence measures using training data. We then train regression forests again to predict the correctness (confidence) of a match by using selected confidence measures. In addition, we present a confidence-based matching cost modulation scheme based on the predicted correctness for improving the robustness and accuracy of various stereo matching algorithms. We apply the proposed scheme to the semi-global matching algorithm to make it robust under unexpected difficulties that can occur in outdoor environments. We verify the proposed confidence measure selection and cost modulation methods through extensive experimentation with various aspects using KITTI and challenging outdoor datasets.

Online multi-object tracking with a single moving camera is a challenging problem as the assumptions of 2D conventional motion models (e.g., first or second order models) in the image coordinate no longer hold because of global camera motion. In this paper, we consider motion context from multiple objects which describes the relative movement between objects and construct a Relative Motion Network (RMN) to factor out the effects of unexpected camera motion for robust tracking. The RMN consists of multiple relative motion models that describe spatial relations between objects, thereby facilitating robust prediction and data association for accurate tracking under arbitrary camera movements. The RMN can be incorporated into various multi-object tracking frameworks and we demonstrate its effectiveness with one tracking framework based on a Bayesian filter. Experiments on benchmark datasets show that online multi-object tracking performance can be better achieved by the proposed method.

Recent achievement of the learning-based classification leads to the noticeable performance improvement in automatic polyp detection. Here, building large good datasets is very crucial for learning a reliable detector. However, it is practically challenging due to the diversity of polyp types, expensive inspection, and labor-intensive labeling tasks. For this reason, the polyp datasets usually tend to be imbalanced, i.e., the number of non-polyp samples is much larger than that of polyp samples, and learning with those imbalanced datasets results in a detector biased toward a non-polyp class. In this paper, we propose a data sampling-based boosting framework to learn an unbiased polyp detector from the imbalanced datasets. In our learning scheme, we learn multiple weak classifiers with the datasets rebalanced by up/down sampling, and generate a polyp detector by combining them. In addition, for enhancing discriminability between polyps and non-polyps that have similar appearances, we propose an effective feature learning method using partial least square analysis, and use it for learning compact and discriminative features. Experimental results using challenging datasets show obvious performance improvement over other detectors. We further prove effectiveness and usefulness of the proposed methods with extensive evaluation.

Online multi-object tracking aims at producing complete tracks of multiple objects using the information accumulated up to the present moment. It still remains a difficult problem in complex scenes, because of frequent occlusion by clutter or other objects, similar appearances of different objects, and other factors. In this paper, we propose a robust online multi-object tracking method that can handle these difficulties effectively. We first propose the tracklet confidence using the detectability and continuity of a tracklet, and formulate a multi-object tracking problem based on the tracklet confidence. The multi-object tracking problem is then solved by associating tracklets in different ways according to their confidence values. Based on this strategy, tracklets sequentially grow with online-provided detections, and fragmented tracklets are linked up with others without any iterative and expensive associations. Here, for reliable association between tracklets and detections, we also propose a novel online learning method using an incremental linear discriminant analysis for discriminating the appearances of objects. By exploiting the proposed learning method, tracklet association can be successfully achieved even under severe occlusion. Experiments with challenging public datasets show distinct performance improvement over other batch and online tracking methods.

Recently, mobile devices such as smart phones and quad-copters are being equipped with inertial measurement units (IMUs) because of advances in micro-electro-mechanical systems technology. This has increased the importance of IMU- camera fusion for vision-based applications. However, ultralow-cost IMUs take much less accurate measurements than low-cost and high-cost IMUs. This uncertainty degrades the accuracy and reliability of IMU-camera calibration, which is the most important step for IMU-camera fusion technology. In this paper, we propose three effective algorithms for robust IMU- camera calibration with uncertain measurements: boundary constraint, adaptive prediction, and angular velocity constraint. These algorithms incorporate a Bayesian filtering framework to estimate calibration parameters more efficiently. The experimental results on both simulation and real data demonstrated the superiority of the proposed algorithms.

In this letter, we propose a robust dynamic point clustering method for detecting moving objects in stereo image sequences, which is essential for collision detection in driver assistance system. If multiple objects with similar motions are located in close proximity, dynamic points from different moving objects may be clustered together when using the position and velocity as clustering criteria. To solve this problem, we apply a geometric constraint between dynamic points using line segments. Based on this constraint, we propose a variable K-nearest neighbor clustering method and three cost functions that are defined between line segments and points. The proposed method is verified experimentally in terms of its accuracy, and comparisons are also made with conventional methods that only utilize the positions and velocities of dynamic points.

In this paper, we consider a multiobject tracking problem in complex scenes. Unlike batch tracking systems using detections of the entire sequence, we propose a novel online multiobject tracking system in order to build tracks sequentially using online provided detections. To track objects robustly even under frequent occlusions, the proposed system consists of three main parts: 1) visual tracking with a novel data association with a track existence probability by associating online detections with the corresponding tracks under partial occlusions; 2) track management to associate terminated tracks for linking tracks fragmented by long-term occlusions; and 3) online model learning to generate discriminative appearance models for successful associations in other two parts. Experimental results using challenging public data sets show the obvious performance improvement of the proposed system, compared with other state-of-the-art tracking systems. Furthermore, extensive performance analysis of the three main parts demonstrates effects and usefulness of the each component for multiobject tracking.

One of the most frequently encountered problems in multiview depth video coding (MDVC) is the relatively low temporal and inter-view correlations when compared to that of multiview color video coding (MCVC). This directly results in degraded performance of the conventional inter-prediction. In this study, we analyze noticeable differences in statistical characteristics between MDVC and MCVC. On the basis of these differences, we propose a new depth-discrepancy compensation (DDC) method with adaptive segment management to alleviate the performance degradation. Experimental results show that the proposed method performs better than the original inter-prediction of JMVC v8.2 in terms of depth bit-savings and synthesized view quality.

When there exists camera and scene motion, the disparity of a pixel temporally varies as time goes on. Such temporal disparity variation (TDV) degrades the performance of spatiotemporal stereo matching. In this paper, we devise a robust similarity measure against TDV, and a suitable optimization technique for the proposed measure. We first design the window-based matching cost to evaluate the similarity between pixels for given disparity and a TDV value. We also present the improved spatiotemporal guided-filter-based aggregation technique to gather match costs with temporal weights. The disparity and TDV maps are then obtained by the global optimization. Here, to handle the large number of labels (disparity levels × TDV levels), we use dual-layer belief propagation that requires less computation and memory while producing comparable results with belief propagation using a single layer. Experimental results show the proposed method yields consistent and accurate disparity maps under the TDV.

In this paper, we propose a novel multi-object tracking method to track unknown number of objects with a single camera system. We design the tracking method via probability hypothesis density (PHD) filtering which considers multiple object states and their observations as random finite sets (RFSs). The PHD filter is capable of rejecting clutters, handling object appearances and disappearances, and estimating the trajectories of multiple objects in a unified framework. Although the PHD filter is robust to cluttered environment, it is vulnerable to missed detections. For this reason, we include local observations in an RFS of observation model. Local observations are locally generated near the individual tracks by using on-line trained local detector. The main purpose of the local observation is to handle the missed detections and to provide identity (label information) to each object in filtering procedure. The experimental results show that the proposed method robustly tracks multiple objects under practical situations.

A new method to up-sample low-resolution depth maps to high quality and high resolution is proposed. Range sensors such as time-of-flight cameras yield low-resolution depth maps and the output includes heavy noise at the edges of objects. Recently, many techniques have been proposed to up-sample low-resolution depth maps. However, there is no effective countermeasure to the edge noise problem. The proposed up-sampling method is based on Markov random fields and addresses this edge noise using newly designed confidence weights. The performance of our method is evaluated using error rate and mean absolute error through comparison with existing methods, and results show that the proposed method outperforms conventional methods.

Conventional local feature-based object recognition methods try to recognize learned 3D objects by using unordered local feature matching followed by the verification. However, the matching between unordered feature sets can be ambiguous and, moreover, it is difficult to deal with general shaped 3D objects in the verification stage. In this paper, we present a new framework for general 3D object recognition, which is based on the invariant local features and their 3D information with stereo cameras. We extend the conventional object recognition framework for stereo cameras. Since the proposed method is based on the stereo vision, it is possible to utilize 3D information of local features visible from two cameras.

International audience

On a large group picture, having all people open their eyes can turn out to be a difficult task for photographers. Therefore, in this paper, we describe an original method to automatically correct closed eyes on everyday pictures. For this aim, we explore the combination possibilities of (1) active shape model (ASM) to detect facial features, such as eyes, nose and head shape, and (2) Poisson editing to clone open eyes seamlessly. To improve the performance of seamless cloning, we suggest a pre-processing method that adjusts skin luminosity between two pictures. A nearest neighbor-based search to find the best suited pair of eyes among a set of donor candidates is also presented. We applied the proposed algorithm on several pictures and obtained very natural results, which demonstrates the validity of our approach.

The point ambiguity owing to the ambiguous local appearances of image points is the one of the main causes making the stereo problem difficult. Under the point ambiguity, local similarity measures are easy to be ambiguous and this results in false matches in ambiguous regions. In this paper, we present the new similarity measure to resolve the point ambiguity problem based on the idea that the distinctiveness, not the interest, is the appropriate criterion for the feature selection under the point ambiguity. The proposed similarity measure named the Distinctive Similarity Measure (DSM) is essentially based on the distinctiveness of image points and the dissimilarity between them, which are both closely related to the local appearances of image points; the distinctiveness of an image point is related to the probability of a mismatch while the dissimilarity is related to the probability of a good match. We verify the efficiency of the proposed DSM by using testbed image sets. Experimental results show that the proposed DSM is very effective and can be easily used for improving the performance of existing stereo methods under the point ambiguity.

Recently, many global stereo methods have achieved good results by modeling a disparity surface as a Markov random field (MRF) and by solving an optimization problem with various techniques. However, most global methods mainly focus on how to minimize conventional cost functions efficiently, although it is more important to define cost functions well to improve performance. In this paper, we propose new symmetric cost functions for global stereo methods. We first present a symmetric data cost function for the likelihood and then propose a symmetric discontinuity cost function for the prior in the MRF model for stereo. In defining cost function, both the reference image and the target image are taken into account to improve performance without modeling half-occluded pixels explicitly and without using color segmentation. The performance improvement of stereo matching due to the proposed symmetric cost functions is verified by applying the proposed symmetric cost functions to the belief propagation (BP) based stereo method. Experimental results for standard testbed images show that the performance of the BP based stereo method is greatly improved by the proposed symmetric cost functions.

In this paper, we propose a fast method for separating reflection components using a single color image. We first propose a specular-free two-band image that is a specularity-invariant color image representation. Reflection components separation is achieved by comparing local ratios at each pixel and making those ratios equal in an iterative framework. The proposed method is very fast and shows reasonable results for textured indoor/outdoor images.

We present a new window-based method for correspondence search using varying support-weights. We adjust the support-weights of the pixels in a given support window based on color similarity and geometric proximity to reduce the image ambiguity. Our method outperforms other local methods on standard stereo benchmarks.

In this paper, we present a new area-based method for visual correspondence search that focuses on the dissimilarity computation. Local and area-based matching methods generally measure the similarity (or dissimilarity) between the image pixels using local support windows. In this approach, an appropriate support window should be selected adaptively for each pixel to make the measure reliable and certain. Finding the optimal support window with an arbitrary shape and size is, however, very difficult and generally known as an NP-hard problem. For this reason, unlike the existing methods that try to find an optimal support window, we adjusted the support-weight of each pixel in a given support window. The adaptive support-weight of a pixel is computed based on the photometric and geometric relationship with the pixel under consideration. Dissimilarity is then computed using the raw matching costs and support-weights of both support windows, and the correspondence is finally selected by the WTA (winner-takes-all) method. The experimental results for the rectified real images show that the proposed method successfully produces piecewise smooth disparity maps while preserving sharp depth discontinuities accurately.

In this paper, we present a new dichromatic-based method for estimating illuminant chromaticity using a single image. In the proposed method, specular pixels are roughly extracted from an input image and projected into the three-dimensional dichromatic line space according to their image chromaticities and dichromatic slopes. Reliable dichromatic lines are then obtained by detecting local maxima in the dichromatic line space. Illuminant chromaticity is finally estimated by finding the intersections of the resulting dichromatic lines. The proposed method has some advantages. The proposed method can deal with highly textured images since it does not require precise color segmentation. In addition, the proposed method does not suffer from nearly parallel dichromatic lines and it can estimate artificial illuminant chromaticity.

For the fast and accurate self-localization of mobile robots, landmarks can be used very efficiently in the complex workspace. In this paper, we propose a simple color landmark model for self-localization and a fast landmark detection and tracking algorithm based on the proposed landmark model. We develop a color landmark with symmetric and repetitive structures, which shows invariant color histogram characteristics under some geometric distortions. Detection and tracking of the model are accomplished by a factored sampling technique in which color similarity is estimated by the color histogram intersection. We also use the color similarity to update the color histogram model of the landmark model for robust tracking under illumination change. We demonstrate the feasibility of the proposed technique through experiments in cluttered indoor environments.

Self-calibration algorithms that use only the information in the image have been actively researched. However, most algorithms require bundle adjustment in the projective reconstruction or in the nonlinear minimization. We propose a practical self-calibration algorithm that only requires a linear projective reconstruction. We overcome the sensitivity of the algorithm due to image noises by adding another constraint on the principal point. Also, we propose a variant of linear auto-calibration algorithm which uses the similar assumption of the work of Pollefeys et al. (1998), based on the property of the absolute quadric. Experimental results using real and synthetic images demonstrate the feasibility of the proposed algorithm.