Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks404 from public data
- Image Segmentation Using Deep Learning: A Survey3,949
A comprehensive review of recent pioneering efforts in semantic and instance segmentation, including convolutional pixel-labeling networks, encoder-decoder architectures, multiscale and pyramid-based approaches, recurrent networks, visual attention models, and generative models in adversarial settings are provided.
- Region Covariance: A Fast Descriptor for Detection and Classification1,481
A fast method for computation of covariances based on integral images, and the performance of the covariance features is superior to other methods, as it is shown, and large rotations and illumination changes are also absorbed by the covariances matrix.
- Underwater scene prior inspired deep underwater image and video enhancement1,262
An underwater image enhancement convolutional neural network model based on underwater scene prior based on underwater scene prior is proposed, called UWCNN, which directly reconstructs the clear latent underwater image, which benefits from the underwater scene prior which can be used to synthesize underwater image training data.
- 1,069
- 889
- 863
- 849
- 826
- The Visual Object Tracking VOT2016 Challenge Results776
The Visual Object Tracking challenge VOT2016 goes beyond its predecessors by introducing a new semi-automatic ground truth bounding box annotation methodology and extending the evaluation system with the no-reset experiment.
- A Novel Performance Evaluation Methodology for Single-Target Trackers667
The requirements are the basis of a new evaluation methodology that aims at a simple and easily interpretable tracker comparison and a fully-annotated dataset with per-frame annotations with several visual attributes, which is the largest benchmark to date.
- Going deeper into action recognition: A survey651
This survey provides a comprehensive review of the notable steps taken towards recognizing human actions, starting with the pioneering methods that use handcrafted representations, and then, navigating into the realm of deep learning based approaches.
- 634
- 550
- 548
- See More, Know More: Unsupervised Video Object Segmentation With Co-Attention Siamese Networks537
This work introduces a novel network, called as CO-attention Siamese Network (COSNet), to address the unsupervised video object segmentation task from a holistic view and proposes a unified and end-to-end trainable framework where different co-att attention variants can be derived for mining the rich context within videos.
- 532
- LightenNet: A Convolutional Neural Network for weakly illuminated image enhancement515
A trainable Convolutional Neural Network is proposed for weakly illuminated image enhancement, namely LightenNet, which takes a weakly illumination image as input and outputs its illumination map that is subsequently used to obtain the enhanced image based on Retinex model.
- 490
- A Self-supervised Approach for Adversarial Robustness385
This paper proposes a self-supervised adversarial training mechanism in the input space that provides significant robustness against the unseen adversarial attacks and can be deployed as a plug-and-play solution to protect a variety of vision systems, as it demonstrates for the case of classification, segmentation and detection.
- 376
- Less Is More: Towards Compact CNNs343
This work shows that, by incorporating sparse constraints into the objective function, it is possible to decimate the number of neurons during the training stage, thus theNumber of parameters and the memory footprint of the neural network are reduced, which is desirable at the test time.
- Ultra-Resolving Face Images by Discriminative Generative Networks333
This work presents a discriminative generative network that can ultra-resolve a very low resolution face image of size \(16 \times 16\) pixels to its \(8\times \) larger version by reconstructing 64 pixels from a single pixel.
- A Survey on Deep Learning Technique for Video Segmentation316
This survey comprehensively review two basic lines of research — generic object segmentation in videos, and video semantic segmentation — by introducing their respective task settings, background concepts, perceived need, development history, and main challenges.
- 313
- DeepTrack: Learning Discriminative Feature Representations Online for Robust Visual Tracking275
This paper presents an efficient and very robust tracking algorithm using a single convolutional neural network for learning effective feature representations of the target object in a purely online manner and introduces a novel truncated structural loss function that maintains as many training samples as possible and reduces the risk of tracking error accumulation.
- Face Super-Resolution Guided by Facial Component Heatmaps248
This paper proposes a method that explicitly incorporates structural information of faces into the face super-resolution process by using a multi-task convolutional neural network (CNN) and achieves superior face hallucination results and outperforms the state-of-the-art.
- 244
- Local Gradients Smoothing: Defense Against Localized Adversarial Attacks230
This work has developed an effective method to estimate noise location in gradient domain and transform those high activation regions caused by adversarial noise in image domain while having minimal effect on the salient object that is important for correct classification.
- 223
- Beyond Local Search: Tracking Objects Everywhere with Instance-Specific Proposals207
This work presents an object tracker that is not limited to a local search window and has ability to probe efficiently the entire frame and provides improved robustness for fast moving objects as well as for ultra lowframerate videos.
- 206
- Performance evaluation of object detection and tracking systems202
This paper presents a set of metrics and algorithms from statistical detection and estimation theory tailored to object detection and tracking tasks using frame-based as well as object-based evaluation paradigms for performance evaluation of object tracking systems.
- Robust Abandoned Object Detection Using Dual Foregrounds199
A pixelwise method that employs dual foregrounds to extract temporally static image regions that indicate objects that were brought into the scene at a subsequent time, such as abandoned and removed items, illegally parked vehicles is presented.
- A comparison of methods for non-rigid 3D shape retrieval175
A new benchmark consisting of 600 non-rigid 3D watertight meshes, which are equally classified into 30 categories, is developed to carry out experiments for 11 different algorithms, whose retrieval accuracies are evaluated using six commonly utilized measures.
- 171
- Zero-Shot Object Detection: Learning to Simultaneously Recognize and Localize Novel Concepts169
This work introduces a new Zero-Shot Detection problem setting, which aims at simultaneously recognizing and locating object instances belonging to novel categories without any training examples, and designs an original loss function that achieves synergy between max-margin class separation and semantic space clustering.
- PartSLIP: Low-Shot Part Segmentation for 3D Point Clouds via Pretrained Image-Language Models168
This paper explores an alternative way for low-shot part segmentation of 3D point clouds by leveraging a pretrained image-language model, GLIP, which achieves superior performance on open-vocabulary 2D detection and a novel 2D-to-3D label lifting algorithm.
- 165
- 164
- Background Modeling and Foreground Detection for Video Surveillance163
This book helps researchers and industry developers understand how to apply background models and foreground detection methods to video surveillance and related areas, such as optical motion capture, multimedia applications, teleconferencing, video editing, and humancomputer interfaces.
- Quadruplet Network with One-Shot Learning for Visual Tracking.162
A new quadruplet deep network is proposed to examine the potential connections among the training instances, aiming to achieve a more powerful representation and is evaluated by model-free tracking-by-detection of objects from a single initial exemplar in several visual object tracking benchmarks.
- InverseForm: A Loss Function for Structured Boundary-Aware Segmentation161
A novel boundary-aware loss term for semantic segmentation using an inverse-transformation network, which efficiently learns the degree of parametric transformations between estimated and target boundaries, which allows consistent and significant performance improvement on segmentation backbone models without increasing their size and computational complexity.
- Traffic congestion estimation using HMM models without vehicle tracking154
An unsupervised, low-latency traffic congestion estimation algorithm that operates on the MPEG video data, and employs Gaussian Mixture Hidden Markov Models (GM-HMM) to detect traffic condition.
- DeepTrack: Learning Discriminative Feature Representations by Convolutional Neural Networks for Visual Tracking153
A model-free tracker that outperforms the existing state-of-the-art algorithms and rarely loses the track of the target object is proposed and a class-specific version of the proposed method that is tailored for tracking of a particular object class such as human faces is introduced.
- 148
- 147
- 145
- 144
- 143
- 140
- Regularization of deep neural networks with spectral dropout130
This work introduces a new approach called 'Spectral Dropout' to improve the generalization ability of deep neural networks by eliminating weak and 'noisy' Fourier domain coefficients of the neural network activations, leading to remarkably better results than the current regularization methods.
- Domain Adaptation by Mixture of Alignments of Second-or Higher-Order Scatter Tensors125
This paper investigates the supervised target scenario for which few labeled target training samples per category exist and proposes an approach to the domain adaptation, dubbed So-HoT, based on the mixture of alignments of second-or higher-order scatter statistics between the source and target domains.
- Human Body Tracking by Adaptive Background Models and Mean-Shift Analysis124
The proposed framework is capable of handling shortcomings of template and correspondence based tracking approaches and provides optimal solutions for several other stages including illumination compensation, skin color detection, shadow removal, morphological filtering, event analysis of a tracking system.
- 123
- Fast Detection of Multiple Objects in Traffic Scenes With a Common Detection Framework122
This paper proposes to detect all the three important objects in a single learning-based detection framework, consisting of a dense feature extractor and detectors of three important classes, and proposes an object subcategorization method as a means of capturing the intraclass variation of objects.
- Indoor Scene Understanding in 2.5/3D for Autonomous Agents: A Survey121
This survey paper provides a comprehensive background to the developed techniques according to a taxonomy based on the scene understanding tasks, and summarizes the performance metrics used for evaluation in different tasks and a quantitative comparison among the recent state-of-the-art techniques.
- Tensor Representations via Kernel Linearization for Action Recognition from 3D Skeletons121
This paper presents two different kernels for action recognition, namely a sequence compatibility kernel that captures the spatio-temporal compatibility of joints in one sequence against those in the other, and a dynamics compatibility kernels that explicitly models the action dynamics of a sequence.
- Learning an Invariant Hilbert Space for Domain Adaptation118
This paper introduces a learning scheme to construct a Hilbert space (i.e., a vector space along its inner product) to address both unsupervised and semi-supervised domain adaptation problems to simultaneously minimizing a notion of domain variance while maximizing a measure of discriminatory power.
- 116
- Face Hallucination with Tiny Unaligned Images by Transformative Discriminative Neural Networks116
An end-to-end transformative discriminative neural network devised for super-resolving unaligned and very small face images with an extreme upscaling factor of 8.5 and significantly outperforms the state-of-the-art.
- Handbook on "Background Modeling and Foreground Detection for Video Surveillance"115
This book helps researchers and industry developers understand how to apply background models and foreground detection methods to video surveillance and related areas, such as optical motion capture, multimedia applications, teleconferencing, video editing, and humancomputer interfaces.
- Bregman Divergences for Infinite Dimensional Covariance Matrices114
This work proposes an approach to computing and comparing Covariance Descriptors (CovDs) in infinite-dimensional spaces by first mapping the original data to a high-dimensional Hilbert space, and only then compute the CovDs.
- A Cascaded Convolutional Neural Network for Single Image Dehazing105
Qualitatively and quantitatively experimental results demonstrate that the proposed method effectively removes haze from such images, and outperforms several state-of-the-art dehazing methods.
- On Generating Transferable Targeted Perturbations104
A new generative approach for highly transferable targeted perturbations (TTP) that not only aligns the global distributions of source and target images, but also matches the local neighbourhood structure between the two domains.
- 102
- 100
- 99
- CLNet: A Compact Latent Network for Fast Adjusting Siamese Trackers98
A compact latent network is proposed to quickly adjust the tracking model to adapt to new scenes and a new training approach based on a diverse sample mining strategy is designed to further improve the discrimination ability of the compact latent network.
- Pushing the Limits of Deep CNNs for Pedestrian Detection97
By reusing the convolutional feature maps of a DCNN model as image features to train an ensemble of boosted decision models, this paper is able to achieve the best reported accuracy without using specially designed learning algorithms.
- Bayesian background modeling for foreground detection97
This work proposes a Bayesian learning method that preserves the multimodality of the background process and is capable of estimating the number of required layers to represent each pixel.
- Trajectory Distance Metric Using Hidden Markov Model Based Representation97
A set of novel distance metrics that use model based representations for trajectories are introduced that determine the similarity of trajectories using the conformity of the corresponding HMM models.
- 96
- Robust Online Visual Tracking with a Single Convolutional Neural Network94
This work introduces a novel truncated structural loss function that maintains as many training samples as possible and reduces the risk of tracking error accumulation, thus drift, by accommodating the uncertainty of the model output.
- Achieving real-time object detection and tracking under extreme conditions93
This survey presents a brief analysis of single camera object detection and tracking methods and gives a comparison of their computational complexities.
- Feature Affinity-Based Pseudo Labeling for Semi-Supervised Person Re-Identification91
This paper proposes a novel feature affinity-based pseudo labeling method with two possible label encodings, and is the first study that employs pseudo-labeling by measuring the affinity of unlabeled samples with the underlying clusters of labeled data samples using the intermediate feature representations from deep networks.
- 87
- 86
- Hyperparameter selection of one-class support vector machine by self-adaptive data shifting85
This paper proposes a fully automatic OCSVM hyperparameter selection method, which requires no tuning of additionalhyperparameter, based on a novel self-adaptive “data shifting” mechanism.
- Video anomaly detection and localization by local motion based joint video representation and OCELM85
To model features of normal video events, the newly-emergent one-class Extreme Learning Machine (OCELM) is introduced as the data description algorithm with a tremendous reduction in training time, which makes the approach easier for model updating and more applicable to fast learning from the rapidly generated surveillance data.
- 85
- 84
- 82
- 82
- 82
- 81
- 79
- 78
- IRISformer: Dense Vision Transformers for Single-Image Inverse Rendering in Indoor Scenes77
This work proposes a transformer architecture to simultaneously estimate depths, normals, spatially-varying albedo, roughness and lighting from a single image of an indoor scene, enabling applications like object insertion and material editing in a single unconstrained real image, with greater photorealism than prior works.
- 76
- 74
- Object Detection and Tracking74
This chapter introduces the reader to main trends and provides taxonomy of popular methods to give an insight to underlying ideas as well as to show their limitations in the hopes of facilitating integration of object detection and tracking for more effective business oriented video analytics.
- 73
- 71
- Learning deep structured network for weakly supervised change detection70
This paper presents a weakly supervised approach that needs only image-level labels to simultaneously detect and localize changes in a pair of images and defines a CRF model to incorporate the local differences and context with the dense connections between individual pixels.
- The Thermal Infrared Visual Object Tracking VOT-TIR2016 Challenge Results68
A significant general improvement of results has been observed, which partly compensate for the more difficult sequences introduced into the dataset in VOT-TIR2016.
- 66
- Pedestrian tracking using thermal infrared imaging66
It was found that infrared images enable better image segmentation, but their tracking performance with current algorithms is poorer, and simple fusion of both types of images has produced some improvement in the segmentation step of the tracking algorithms.
- 66
- Video Representation Learning Using Discriminative Pooling65
This work proposes discriminative pooling, based on the notion that among the deep features generated on all short clips, there is at least one that characterizes the action, and learns a (nonlinear) hyperplane that separates this unknown, yet discrim inative, feature from the rest.
- Simultaneous Stereo Video Deblurring and Scene Flow Estimation65
This paper proposes a novel approach to deblurring from stereo videos that exploits the piece-wise planar assumption about the scene and leverage the scene flow information todeblur the image and achieves significant improvement in flow estimation and removing the blur effect over the state-of theart methods.
- 63
- A Deeper Look at Power Normalizations62
A novel layer is introduced by introducing a novel layer that implements PN for non-linear pooling of feature maps by using a kernel formulation, which combines the feature vectors and their respective spatial locations in the feature maps produced by the last convolutional layer of CNN.
- 60
- 59
- Zero-Shot Object Detection: Joint Recognition and Localization of Novel Concepts58
An integrated solution to the ZSD problem that jointly models the complex interplay between visual and semantic domain information and an end-to-end trainable deep network for ZSD that effectively overcomes the noise in the unsupervised semantic descriptions.
- 56
- X-Align++: cross-modal cross-view alignment for Bird’s-eye-view segmentation55
This paper proposes X-Align, a novel end-to-end cross-modal and cross-view learning framework for BEV segmentation consisting of a novel Cross-Modal Feature Alignment (X-FA) loss, and provides extensive ablation studies to demonstrate the effectiveness of the individual components.
- 55
- Feature mask network for person re-identification54
A Feature Mask Network (FMN) is proposed that takes advantage of ResNet high-level features to predict a feature map mask and then imposes it on the low- level features to dynamically re-weight different object parts for a complementary feature representation.
- 54
- 53
- 53
- Depth Estimation and Blur Removal from a Single Out-of-focus Image53
This paper presents a depth estimation method that leverages rich representations learned from cascaded convolutional and fully connected neural networks operating on a patch-pooled set of feature maps and substantially improves depth accuracy over the state-of-the-art alternatives.
- X3KD: Knowledge Distillation Across Modalities, Tasks and Stages for Multi-Camera 3D Object Detection52
This paper introduces X3KD, a comprehensive knowledge distillation framework across different modalities, tasks, and stages for multi-camera 3DOD and proposes cross-task distillation from an instance segmentation teacher (X-IS) in the PV feature extraction stage providing supervision without ambiguous error backpropagation through the view transformation.
- Adaptive Siamese Tracking with a Compact Latent Network52
This article provides an intuitive viewing to simplify the Siamese-based trackers by converting the tracking task to a classification, and presents a statistics-based compact latent feature for fast adjustment by efficiently extracting the sequence-specific information.
- Museum Exhibit Identification Challenge for the Supervised Domain Adaptation and Beyond52
An open problem of artwork identification is studied and a new dataset dubbed Open Museum Identification Challenge (Open MIC), which contains photos of exhibits captured in 10 distinct exhibition spaces of several museums is proposed to stimulate research in domain adaptation, egocentric recognition and few-shot learning.
- 51
- AuxAdapt: Stable and Efficient Test-Time Adaptation for Temporally Consistent Video Semantic Segmentation50
An efficient, intuitive, and unsupervised online adaptation method, AuxAdapt, for improving the temporal consistency of most neural network models, which does not require optical flow and only takes one pass of the video.
- 50
- Super-Trajectory for Video Segmentation50
A novel semi-supervised video segmentation approach based on an efficient video representation, called as “super-trajectory”, which is capable of distinguishing the target objects from complex backgrounds and even reidentifying them after occlusions.
- 50
- 49
- 49
- Ordered Pooling of Optical Flow Sequences for Action Recognition47
This paper introduces a novel ordered representation of consecutive optical flow frames as an alternative and argues that this representation captures the action dynamics more efficiently than RGB frames, and provides intuitions on why such a representation is better for action recognition.
- 47
- 47
- Skip-Attention: Improving Vision Transformers by Paying Less Attention46
SkipAt, a method to reuse self-attention computation from preceding layers to approximate attention at one or more subsequent layers, is proposed, which outperforms the baseline transformer's performance while running computationally faster.
- Imposing Consistency for Optical Flow Estimation46
It is demonstrated that the consistency strategies applied to a strong baseline network model using the original datasets and labels provide further improvements, attaining the state-of-the-art results on the KITTI-2015 scene flow benchmark in the non-stereo category.
- 46
- 46
- Sort-free Gaussian Splatting via Weighted Sum Rendering45
Weighted Sum Rendering is proposed, which approximates alpha blending with weighted sums, thereby removing the need for sorting, and which simplifies implementation, delivers superior performance, and eliminates the "popping"artifacts caused by sorting.
- Component Attention Guided Face Super-Resolution Network: CAGFace45
A fully convolutional multi-stage neural network for 4× super-resolution for face images that implicitly impose facial component-wise attention maps using a segmentation network to allow the network to focus on face-inherent patterns.
- 45
- Hallucinating Unaligned Face Images by Multiscale Transformative Discriminative Networks44
An end-to-end multiscale transformative discriminative neural network devised for super-resolving unaligned and very small face images of different resolutions ranging from 16 to 32 pixels in a unified framework that significantly outperforms the state-of-the-art face hallucination networks.
- 43
- Deblur and deep depth from single defocus image42
This article presents a depth estimation method that leverages rich representations learned from cascaded convolutional and fully connected neural networks operating on a patch-pooled set of feature maps and substantially improves depth accuracy over the state-of-the-art alternatives.
- Real-time video object segmentation for MPEG-encoded video sequences42
A real-time object segmentation method for MPEG encoded video that exploits the macro-block structure of the encoded video to decrease the spatial resolution of the processed data, which exponentially reduces the computational load.
- Advances in Visual Computing41
A food recognition algorithm based on sparse representation that learns overcomplete dictionaries from local descriptors including texture and color features that are extracted from food image patches that comprises 18 classes.
- 40
- 40
- Selective Video Object Cutout39
A pyramid histogram-based confidence map that incorporates structure information into appearance statistics and employs an efficient measure of uncertainty propagation using local classifiers to determine the image regions, where the object labels might be ambiguous.
- 39
- 38
- 37
- 37
- 37
- 37
- 37
- 36
- Robust visual tracking with channel attention and focal loss36
This work introduces channel attention and focal loss into the network design to enhance feature representation learning and demonstrates that the enhanced tracker achieves significant performance improvement while running at a real-time frame-rate of 66 fps.
- 36
- 36
- 35
- 35
- 34
- MAMo: Leveraging Memory and Attention for Monocular Video Depth Estimation33
MAMo is proposed, a novel memory and attention framework for monocular video depth estimation that consistently improves monocular depth estimation networks and sets new state-of-the-art (SOTA) accuracy.
- 33
- Factorized Inverse Path Tracing for Efficient and Accurate Material-Lighting Estimation32
The Factorized Inverse Path Tracing (FIPT) addresses challenges by using a factored light transport formulation and finds emitters driven by rendering errors, and enables accurate material and lighting optimization faster than previous work, and is more effective at resolving ambiguities.
- 32
- 32
- Self-Supervised Geometric Correspondence for Category-Level 6D Object Pose Estimation in the Wild31
Surprisingly, this method, without any human annotations or simulators, can achieve on-par or even better performance than previous supervised or semi-supervised methods on in-the-wild images.
- 31
- A new approach for in-vehicle camera traffic sign detection and recognition31
This paper discusses theoretical foundations and a practical realization of a circular traffic sign detection and recognition system operating on board of a vehicle and adopts a novel AdaBoost algorithm to learn a sign similarity measure from example image pairs labeled either “same” or “different”.
- Consistency and Diversity Induced Human Motion Segmentation29
A novel Consistency and Diversity induced human Motion Segmentation (CDMS) algorithm that factorizes the source and target data into distinct multi-layer feature spaces, in which transfer subspace learning is conducted on different layers to capture multi-level information.
- 29
- 29
- Convolutional neural net bagging for online visual tracking29
Experiments demonstrate that the proposed CNN tracker outperforms the state-of-the-art methods on three recent benchmarks, which illustrates the superiority of the feature representations learned by the purely online bagging framework.
- 29
- EGA-Depth: Efficient Guided Attention for Self-Supervised Multi-Camera Depth Estimation27
A novel guided attention architecture, EGA-Depth, which can improve both the efficiency and accuracy of self-supervised multi-camera depth estimation by using its perspective view as the query to cross-reference its neighboring views to derive informative features for this camera view.
- 27
- 26
- 26
- 26
- Cost-sensitive joint feature and dictionary learning for face recognition25
A cost-sensitive joint feature and dictionary learning (CS-JFDL) method, in which the features are concurrently learned with the dictionaries, which demonstrates that the proposed algorithm is better than many state-of-the-art methods.
- Identity-Preserving Face Recovery from Stylized Portraits25
An Identity-preserving Face Recovery from Portraits method that utilizes a Style Removal network (SRN) and a Discriminative Network (DN) that can recover high-quality photorealistic faces from unaligned portraits while preserving the identity of the face in an image is developed.
- 25
- OCAI: Improving Optical Flow Estimation by Occlusion and Consistency Aware Interpolation24
This work proposes OCAI, a method that supports robust frame interpolation by generating intermediate video frames along-side optical flows in between and introduces a teacher-student style semi-supervised learning method on top of the interpolated frames.
- HexaGen3D: StableDiffusion is just one step away from Fast and Diverse Text-to-3D Generation24
HexaGen3D does not require per-sample optimization, and can infer high-quality and diverse objects from textual prompts in 7 seconds, offering significantly better quality-to-latency trade-offs when comparing to existing approaches.
- X-Distill: Improving Self-Supervised Monocular Depth via Cross-Task Distillation24
A novel method to improve the self-supervised training of monocular depth via cross-task knowledge distillation from semantic segmentation to depth estimation, which significantly improves the depth estimation accuracy and outperforms the state of the art.
- Sketch-specific data augmentation for freehand sketch recognition24
A novel sketch-specific data augmentation (SSDA) method that leverages the quantity and quality of the sketches automatically and can be integrated with any convolutional neural networks, it has a distinct advantage over the existing methods.
- 24
- Clustering Variable Length Sequences by Eigenvector Decomposition Using HMM24
This work presents a novel clustering method using HMM parameter space and eigenvector decomposition that can cluster both constant and variable length sequences without requiring normalization of data.
- 23
- Simple and Efficient Architectures for Semantic Segmentation23
It is demonstrated that a simple encoder-decoder architecture with a ResNet-like backbone and a small multi-scale head, performs on-par or better than complex semantic segmentation architectures such as HRNet, FANet and DDRNets.
- Identity Enhanced Residual Image Denoising23
The proposed network produces remarkably higher numerical accuracy and better visual image quality than the classical state-of-the-art and CNN algorithms when being evaluated on the three conventional benchmark and three real-world datasets.
- 23
- 23
- 22
- 22
- DistractFlow: Improving Optical Flow Estimation via Realistic Distractions and Pseudo-Labeling21
A novel data augmentation approach for training optical flow estimation models by introducing realistic distractions to the input frames by combining one of the frames in the pair with a distractor image depicting a similar domain, which allows for inducing visual perturbations congruent with natural objects and scenes.
- 4D Panoptic Segmentation as Invariant and Equivariant Field Prediction21
This paper develops rotation-equivariant neural networks for 4D panoptic segmentation and sets the new state-of-the-art performance and achieves 1st place on the SemanticKITTI 4D Panoptic Segmentation leaderboard.
- Stylized Adversarial Defense21
This work uses the style and content information of the target sample from another class, alongside its class-boundary information, to create adversarial perturbations, and proposes a max-margin adversarial training approach that minimizes the distance between source image and its adversary and maximizes thedistance between the adversary and the target image.
- 21
- The VOT2013 challenge: overview and additional results21
An overview of the VOT2013 challenge is provided, its main results are pointed out, and the additional previously unpublished experiments and results are documented.
- Clockwork Diffusion: Efficient Generation With Model-Step Distillation20
This work proposes Clockwork Diffusion, a method that periodically reuses computation from preceding denoising steps to approximate low-res feature maps at one or more subsequent steps and demonstrates that Clockwork leads to compa-rable or improved perceptual scores with drastically re-duced computational complexity.
- HS3: Learning with Proper Task Complexity in Hierarchically Supervised Semantic Segmentation20
This paper proposes Hierarchically Supervised Semantic Segmentation (HS3), a training scheme that supervises intermediate layers in a segmentation network to learn meaningful representations by varying task complexity, and devise a fusion framework, HS3-Fuse, to aggregate the hierarchical features generated by these layers.
- Panoptic, Instance and Semantic Relations: A Relational Context Encoder to Enhance Panoptic Segmentation19
This paper presents a novel framework to integrate both semantic and instance contexts for panoptic segmentation, and proposes a novel Panoptic, Instance, and Semantic Relations (PISR) module to exploit such contexts.
- Perceptual Consistency in Video Segmentation19
A novel perceptual consistency perspective on video semantic segmentation is presented, which can capture both temporal consistency and pixel-wise correctness and be used as a regularizer during the training of segmentation models, which leads to more temporally consistent video segmentation while maintaining accuracy.
- 19
- Distinctive action sketch for human action recognition19
This paper proposes an approach to discover the most distinctive sketches for action recognition, and presents four kinds of sketch pooling methods to get a uniform representation for action videos.
- 19
- 18
- Learning target-aware correlation filters for visual tracking18
The TACF formulation is proposed, which proposes an optimization strategy based on the Preconditioned Conjugate Gradient method for efficient filter learning and achieves state-of-the-art performance on OTB100 and the top rank in EAO on the VOT2016 challenge.
- 18
- Robust and real-time deep tracking via multi-scale domain adaptation18
This paper proposes to transfer the feature for image classification to the visual tracking domain via convolutional channel reductions via convolved channel reductions and illustrates the state-of-the-art accuracies in the experiment involving two well-adopted benchmarks with more than 100 test videos.
- 18
- 18
- Collaborative tracking of objects in EPTZ cameras18
This paper proposes a new joint tracking paradigm for the multiple stream electronic pan-tilt-zoom (EPTZ) cameras, capable of transmitting a low resolution thumbnail (LRT) image of the whole field of view as well as a high-resolution cropped (HRC) image for the target region.
- 18
- Identity-Preserving Face Recovery from Portraits17
A new Identity-preserving Face Recovery from Portraits (IFRP) to recover latent photorealistic faces from unaligned stylized portraits by embedding spatial transformer networks into the SRN, which can compensate for misalignments of stylized faces automatically and output aligned realistic face images.
- Depth Map Completion by Jointly Exploiting Blurry Color Images and Sparse Depth Maps17
This paper proposes to tackle the problem of depth map completion by jointly exploiting the blurry color image sequences and the sparse depth map measurements, and presents an energy minimization based formulation to simultaneously complete the depth maps, estimate the scene flow and deblur the color images.
- 17
- Tracklet clustering for robust multiple object tracking using distance dependent Chinese restaurant processes17
This work presents a tracklet clustering approach using distance dependent Chinese restaurant processes (ddCRPs), which employ a two-level robust object tracker, and can cluster complex tracklets by only computing similarities between them.
- 17
- Saliency Integration: An Arbitrator Model16
This paper proposes an arbitrator model (AM) for saliency integration that incorporates the consensus of multiple saliency models and the external knowledge into a reference map to effectively rectify the misleading by candidate models and derives a Bayesian integration framework to reconcile theSaliency models of varying expertise and the reference map.
- CustomKD: Customizing Large Vision Foundation for Edge Model Improvement via Knowledge Distillation15
Beyond providing well-generalized original knowledge from teachers, CustomKD aligns the features of teachers to those of students, making it easy for students to understand and overcome the large model discrepancy overall.
- 15
- Deep Learning for Visual Understanding: Part 2 [From the Guest Editors]15
Computer vision aims to make computers understand the seen data in the same way that human vision does, and incorporates several scientific disciplines such as signal processing, machine learning, applied mathematics, sensing, geometry, optimization, statistics, and data sciences.
- QuickSRNet: Plain Single-Image Super-Resolution Architecture for Faster Inference on Mobile Platforms14
This work proposes QuickSRNet, a simple yet effective architecture that provides better accuracy-to-latency trade-offs than existing neural architectures for single-image super-resolution and presents training tricks to speed up existing residual-based super- resolution architectures while maintaining robustness to quantization.
- 14
- 14
- No fuss metric learning, a Hilbert space scenario14
A kernel version of the recently introduced keep it simple and straightforward metric learning method is devised, hence adding a novel dimension to its applicability in scenarios where input data is non-linearly distributed, and proposes two techniques towards projecting on the positive cone in a reproducing kernel Hilbert space.
- Face Destylization14
A novel Face Destylization Neural Network (FDNN) is proposed to restore the latent photo- realistic faces from the stylized ones and can recover faces from stylized portraits and real paintings for which the stylization data was unavailable during the training phase.
- Sparse Dictionaries for Semantic Segmentation14
A CRF model is proposed that incorporates a DSDL based top-down cost for semantic segmentation and it is shown that the new CRF energy can be minimized using existing efficient discrete optimization techniques.
- Advances in Visual Computing14
This paper proposes several features for the analysis of color symmetry in skin lesions and assesses their performance under a wide variety of system configurations, obtaining very high detection scores by only using color symmetry features.
- Road extraction by point-wise Gaussian models14
A Gaussain model based approach for robust and automatic extraction of roads from very low-resolution satellite imagery by evaluating the responses from a quadruple orthogonal line filter set.
- 13
- Recovering Faces From Portraits with Auxiliary Facial Attributes13
This work proposes an Attribute-guided Face Recovery from Portraits (AFRP) that utilizes a Face Recovery Network (FRN) and a Discriminative Network (DN), and demonstrates that the method can recover photorealistic identity-preserving faces with desired attributes from unseen stylized portraits, artistic paintings, and hand-drawn sketches.
- 13
- 13
- 13
- Fast distance transform computation using dual scan line propagation13
Two fast algorithms that approximate the distance transformation of 2D binary images by using either multi-directional dual scan line propagation or wave propagation methods to extract the minimum distances with no explicit distance computation.
- Neodragon: Mobile Video Generation using Diffusion Transformer12
This work introduces Neodragon, a text-to-video system capable of generating 2s (49 frames @24 fps) videos at the 640x1024 resolution directly on a Qualcomm Hexagon NPU in a record 6.7s (7 FPS), the first to have been specifically optimised for mobile hardware to achieve efficient and high-fidelity video synthesis.
- DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions12
This paper enhances a baseline convolutional depth completion model by applying attention to 2D features in the bottleneck and skip connections and proposes normalization techniques to process the point cloud, which improves learning and leads to better accuracy than directly using point transformers off the shelf.
- Indoor Scene Understanding in 2.5/3D: A Survey.12
This survey paper provides a comprehensive background to this research topic, followed by popular 3D data representations and a comparative analysis of available datasets, and summarizes the performance metrics used for evaluation in different tasks and a quantitative comparison among the recent state of theart techniques.
- 12
- 12
- 12
- 12
- LoRA-X: Bridging Foundation Models with Training-Free Cross-Model Adaptation11
A new adapter, Cross-Model Low-Rank Adaptation (LoRA-X), is introduced, which enables the training-free transfer of LoRA parameters across source and target models, eliminating the need for original or synthetic training data.
- Planar Gaussian Splatting11
Planar Gaussian Splatting achieves state-of-the-art performance in 3D planar reconstruction without requiring either 3D plane labels or depth supervision, and maintains its performance across datasets thanks to its neural rendering and scene-specific optimization mechanism.
- PosSAM: Panoptic Open-vocabulary Segment Anything11
An open-vocabulary panoptic segmentation model that effectively unifies the strengths of the Segment Anything Model with the vision-language CLIP model in an end-to-end framework and proposes a novel Local Discriminative Pooling (LDP) module leveraging class-agnostic SAM and class-aware CLIP features for unbiased open-vocabulary classification.
- 11
- 11
- 11
- Integrated deep and shallow networks for salient object detection11
This paper takes results of unsupervised saliency and normalized color images as inputs, and directly learns an end-to-end mapping between inputs and the corresponding saliency maps to obtain a spatially consistent saliency map with sharp object boundaries.
- Combined Internal and External Category-Specific Image Denoising11
A category-specific image denoising algorithm that exploits patch similarity within theinput image and between the input image and an external dataset that produces superior results compared with state-of-the-art denoised methods both qualitatively and quantitatively.
- 11
- 11
- Multi-Camera Surveillance: Object-Based Summarization Approach11
An automatic object tracking and video summarization method for multicamera systems with a large number of non-overlapping field-of-view cameras is explained, which enables annotation of video segments, and extraction of content semantics for further analysis and summarization.
- DIFT: Dynamic Iterative Field Transforms for Memory Efficient Optical Flow10
This paper introduces a lightweight low-latency and memory-efficient model, Dynamic Iterative Field Transforms (DIFT), for optical flow estimation feasible for edge applications such as mobile, XR, micro UAVs, robotics and cameras, and proposes a memory efficient solution for cost volume processing to reduce peak memory.
- 10
- Harmonic Variance: A Novel Measure for In-focus Segmentation10
The harmonic mean of variances is computed from the statistical properties of the image in its bandpass filtered versions and incorporated into a graph Laplacian spectrum based segmentation framework in order to accurately align the in-focus measure responses on the underlying image structure.
- PADRe: A Unifying Polynomial Attention Drop-in Replacement for Efficient Vision Transformer9
PADRe runs significantly faster than the conventional self-attention while maintaining similar accuracy when substituting self-attention in the transformer models, and is assessed as a drop-in replacement for self-attention across diverse computer vision tasks.
- EdgeRelight360: Text-Conditioned 360-Degree HDR Image Generation for Real-Time On-Device Video Portrait Relighting9
The method proposes a diffusion-based text-to-360-degree image generation in the HDR domain, taking advantage of the HDR10 standard, which facilitates the generation of high-quality, realistic lighting conditions from textual descriptions, offering flexibility and control in portrait video relighting task.
- 9
- Special issue on dynamic textures in video9
By segmenting and excluding dynamic textures, the robustness of the moving object detection and action identification can be improved and practical applications include detection of certain types of dynamic textures such as fire and smoke, realistic rendering and compact visualization of dynamicTextures, and efficient retrieval of video in multimedia databases.
- Learning on Manifolds9
Two Riemannian manifolds including the set of affine transformations and covariance matrices are elaborated and their favorable applications in distance computation, motion estimation, object detection and recognition problems are demonstrated.
- RoCA: Robust Cross-Domain End-to-End Autonomous Driving8
This paper proposes RoCA, a novel framework for robust cross-domain E2E autonomous driving that achieves strong domain generalization and adaptation performance and enables robust adaptation on new target domains, significantly outperforming direct finetuning.
- Dynamic Iterative Refinement for Efficient 3D Hand Pose Estimation8
A tiny deep neural network of which partial layers are recursively exploited for refining its previous estimations, which consistently outperforms state-of-the-art 2D/3D hand pose estimation approaches in terms of both accuracy and efficiency for widely used benchmarks.
- Distorting Neural Representations to Generate Highly Transferable Adversarial Examples.8
This work proposes a novel approach to create adversarial examples that can broadly fool different networks on multiple tasks, based on the following intuition: "Deep features are highly generalizable and show excellent performance across different tasks, therefore an ideal attack must create maximum distortions in the feature space to realize highly transferable examples".
- 8
- 8
- 8
- 8
- 8
- 8
- Video object segmentation8
A unique framework that combines various disciplines of image and video processing techniques ranged from noise filtering to data clustering is developed, and it addresses a number of challenging issues associated with computational complexity, accuracy, generality, and robustness.
- 7
- ToSA: Token Selective Attention for Efficient Vision Transformers7
A novel token selective attention approach, ToSA, which can identify tokens that need to be attended as well as those that can skip a transformer layer, and which can significantly reduce computation costs while maintaining accuracy on the ImageNet classification benchmark.
- 7
- 7
- Deep Learning for Visual Understanding [From the Guest Editors]7
In the past decade, there has been a transformative and permanent revolution in computer vision cultivated by the reinvigorated adoption of deep learning for visual understanding tasks, driven by the increasing availability of large annotated data sets, efficient training techniques, and faster computational platforms.
- 7
- Model-Free Multiple Object Tracking with Shared Proposals7
A novel model-free framework that incorporates generic category-independent object proposals without the need to pretrain any object detectors is presented, which significantly improves the search efficiency in comparison to the traditional dense sampling approach.
- Weakly Supervised Change Detection in a Pair of Images.7
This paper employs a deep neural network with DAG topology to learn patterns of change from image-level labeled training data, and applies a constrained mean-field algorithm to estimate the pixel-level labels, which enables imposing global constraints on the observed foreground probability mass function.
- Biomechanical model-based 4DCT simulation7
A novel heterogeneous design can leverage the advantages of both continuum mechanics and mass–spring–damper system in the way that the lung deformation is computed in very high accuracy while the deformation of the rest CT scans is simulated using finite difference analysis.
- Detecting and Tracking Unknown Number of Objects with Dirichlet Process Mixture Models and Markov Random Fields7
An object tracking framework that employs Dirichlet Process Mixture Models (DPMMs) in a multiple hypothesis tracker that group tracked hypotheses to combine possible parts of an object as one is presented.
- 7
- 7
- Fast image registration via joint gradient maximization: application to multimodal data7
This work presents a computationally inexpensive method for multi-modal image registration that employs a joint gradient similarity function that is applied only to a set high spatial gradient pixels.
- Generative Scenario Rollouts for End-to-End Autonomous Driving6
Generative Scenario Rollouts (GeRo), a plug-and-play framework for VLA models that jointly performs planning and generation of language-grounded future traffic scenes through an autoregressive rollout strategy, achieves state-of-the-art closed-loop and open-loop performance and demonstrates strong zero-shot robustness.
- FLoC: Facility Location-Based Efficient Visual Token Compression for Long Video Understanding6
FLoC is an efficient visual token compression framework based on the facility location function, a principled approach that swiftly selects a compact yet highly representative and diverse subset of visual tokens within a predefined budget on the number of visual tokens.
- 6
- MoViE: Mobile Diffusion for Video Editing6
A series of optimizations that render mobile video editing feasible are introduced, first optimize its architecture and incorporate a lightweight autoencoder, resulting in a threefold on-device speedup and a novel adversarial distillation scheme which preserves the controllability of the editing process.
- 6
- Neural 5G Indoor Localization with IMU Supervision6
This work uses pseudo-labels during deployment to perform similarly to fully-supervised, but requires much less effort to deploy, and proposes practical algorithms for IMU double integration and training of the localization system.
- Transadapt: A Transformative Framework for Online Test Time Adaptive Semantic Segmentation6
This work proposes TransAdapt, a framework that uses transformer and input transformations to improve segmentation performance in online settings and pre-train a transformer-based module on a segmentation network that transforms unsupervised segmentation output to a more reliable supervised output, without requiring test-time online training.
- Beyond feature integration: a coarse-to-fine framework for cascade correlation tracking6
This paper presents a coarse-to-fine framework for cascade correlation tracking (CCT), which decomposes a complicated tracker into two low-complexity modules, a coarse tracker and a refined tracker working in a coarse-to-fine manner.
- 6
- 6
- 6
- 6
- Designing a Boosted Classifier on Riemannian Manifolds6
This chapter describes a boosted classification approach that incorporates the a priori knowledge of the geometry of the Riemannian space into a rejection cascade and applied to single image human detection task.
- Reshuffling: a fast algorithm for filtering with arbitrary kernels6
A novel method to accelerate the application of linear filters that have multiple identical coefficients on arbitrary kernels is presented, and results prove that the reshuffling is superior to the conventional approach.
- 6
- Zero-Shot Adaptation of Parameter-Efficient Fine-Tuning in Diffusion Models5
ProLoRA transfers pre-trained low-rank adjustments from a source to a target model without additional training data, enabling zero-shot adaptation of parameter-efficient fine-tuning in text-to-image diffusion models.
- Efficient neural supersampling on a novel gaming dataset5
This work introduces a novel neural algorithm for super-sampling rendered content that is 4× more efficient than existing methods while maintaining the same level of accuracy.
- Test-time Adaptation vs. Training-time Generalization: A Case Study in Human Instance Segmentation using Keypoints Estimation5
It is shown that in the absence of a significant domain shift, TTA may hurt and TTG show only a small gain in performance, whereas for a large domain shift), TTA gains are smaller and dependent on the heuristics used, while TTG gains are larger and robust to architectural choices.
- 5
- 5
- Regression on Lie Groups and Its Application to Affine Motion Tracking5
This chapter forms a regression model on the corresponding Lie algebra that minimizes a first order approximation to the geodesic error and demonstrates the method on affine motions, however, it generalizes to any matrix Lie group transformations.
- Special issue on car navigation and vehicle systems5
This special issue brings together several contributions toward achieving visual intelligence in autonomous navigation by examining the optical flow observed by a camera, mounted in a car and looking forward.
- 5
- 5
- 5
- Automatic image segmentation by wave propagation5
A level set based region growing method for automatic partitioning of color images into segments that prevents leakage and imposes compactness on the region without over-smoothing its boundary and can deal with sharp corners and changes in topology.
- 5
- 5
- A Comparative analysis of Layer-wise Representational Capacity in AR and Diffusion LLMs4
First layer-and token-wise representational analysis comparing native dLLMs, native AR models, and AR-initialized dLLMs finds that diffusion objectives result in different, more hierarchical abstraction with substantial early-layer redundancy and reduced recency bias, while AR objectives produce tightly coupled, depth-dependent representations.
- H3O: Hyper-Efficient 3D Occupancy Prediction with Heterogeneous Supervision4
This paper presents a novel 3D occupancy prediction approach, H30, which features highly efficient architecture designs that incur a significantly lower computational cost as compared to the current state-of-the-art methods.
- 4
- 4
- Introduction to the Special Issue on Tensor Decomposition for Signal Processing and Machine Learning4
- 4
- Towards Purely Unsupervised Disentanglement of Appearance and Shape for Person Images Generation4
An encoder-decoder-like network to extract both the shape and appearance features from input images at the same time, and train the parameters by three losses: feature adversarial loss, color consistency loss and reconstruction loss, which achieves clean disentanglement.
- Cascade residuals guided nonlinear dictionary learning4
This paper presents a two-pass multi-resolution cascade framework for dictionary learning and sparse coding, which allows collaborative reconstructions at different resolutions using only the same dimensional dictionary atoms.
- Object-Aware Dictionary Learning with Deep Features4
This work presents an object-aware dictionary learning framework that systematically incorporates region proposals and deep features in order to improve the discriminative power of the combined classifier and proposes an iterative solver.
- 4
- Biomechanical Simulation of Lung Deformation from One CT Scan4
A biomechanical model based simulation method for examining the patient lung deformation induced by respiratory motion, given only one CT scan input, and introduces robust algorithms to segment out the diaphragm control points and spine regions to carefully define the boundary conditions and loads.
- SAR despeckling by sparse reconstruction on affinity nets (SRAN)4
This paper presents a new approach for multiplicative noise removal in SAR images based on sparse coding by dictionary learning and collaborative filtering, and demonstrates superior despeckle filtering performance.
- Needle picking: a sampling based track-before-detection method for small targets4
A computationally efficient track-before-detect algorithm that achieves more than 50% true detection at 10-6 false alarm rate for pixel sized unknown number of multiple targets when the signal-to-noise ratio is less than 7dB is presented.
- 4
- Estimation of contour motion and deformation for nonrigid object tracking4
This work solves the nonrigid contour tracking problem by decomposing it into three subproblems: motion estimation, deformation estimation, and shape regulation, and uses a shape prior model to constrain the deformed contour.
- CoReDiT: Spatial Coherence-Guided Token Pruning and Reconstruction for Efficient Diffusion Transformers3
CoReDiT, a structured token pruning framework for DiTs across vision tasks, uses a linear-time spatial coherence score to estimate local redundancy in the latent token lattice and skips high coherence (redundant) tokens in self-attention to maintain a dense representation.
- MobileWan: Closing the Quality Gap for Mobile Video Diffusion3
MobileWan becomes the first 5B-scale video diffusion model deployable on a commercial mobile device and proposes a learnable attention head pruning method based on binary per-head gates optimized end-to-end using a noise-biased sparsity objective and distillation-based finetuning.
- CA-LoRA: Concept-Aware LoRA for Domain-Aligned Segmentation Dataset Generation3
This paper proposes Concept-Aware LoRA (CA-LoRA), a novel fine-tuning approach that selectively identifies and updates only the weights associated with necessary concepts for domain alignment while preserving the pretrained knowledge of the T2I model to produce informative samples.
- DySS: Dynamic Queries and State-Space Learning for Efficient 3D Object Detection from Multi-Camera Videos3
DySS leverages a state-space model (SSM) to sequentially process the sampled features over time steps and dynamically update the queries via merge, remove, and split operations, which help maintain a useful, lean set of detection queries throughout the network.
- FALO: Fast and Accurate LiDAR 3D Object Detection on Resource-Constrained Devices3
This paper proposes FALO, a hardware-friendly approach to LiDAR 3D detection, which offers both state-of-the-art (SOTA) detection accuracy and fast inference speed and can readily deploy on compact, embedded devices.
- 3
- Segmentation-Free Guidance for Text-to-Image Diffusion Models3
S segmentation-free guidance, a novel method designed for text-to-image diffusion models like Stable Diffusion, uses the diffusion model itself as an implied segmentation network, hence named segmentation-free guidance, to dynamically adjust the negative prompt for each patch of the generated image, based on the patch’s relevance to concepts in the prompt.
- SciFlow: Empowering Lightweight Optical Flow Models with Self-Cleaning Iterations3
Two synergistic techniques, Self-Cleaning Iteration (SCI) and Regression Focal Loss (RFL), designed to enhance the capabilities of optical flow models, with a focus on addressing optical flow regression ambiguities, prove particularly effective in mitigating error propagation.
- Machine learning for big visual analysis3
This special issue aims to demonstrate the contribution of machine learning techniques to the research and development of big visual data analysis by evaluating the performance of convolutional neural networks on large-scale image classification.
- Breaking video into pieces for action recognition3
This work partitions the input video into small clips and formulate action recognition as a joint decision-making task, and builds an aggregated representation for the original full-length video by integrating the elements of the DVM.
- 3
- 3
- 3
- Computationally efficient histogram extraction for rectangular image regions3
This work presents a novel method, which is computationally superior and makes it possible to employ even an exhaustive search process in real-time, which was impractical before, and enables the description of higher level histogram features.
- Object Segmentation of Color Video Sequences3
A video segmentation algorithm that accurately finds object boundaries, and does not require any user assistance is presented, which proves the effectiveness of the algorithm.
- Disparity estimation by patch matching3
A stripe-mesh type image representation scheme and a disparity estimation algo-rithm based on the stripe-mesh structure are proposed that acquires the structure of the scene by estimating the disparities within the mesh patches, using a modi(cid:12)ed exhaustive search algo-rithm.
- 2
- 2
- Online Adaptive Personalization for Face Anti-Spoofing2
Through experimental evaluation on the SiW dataset, it is shown that OAP improves recognition performance of existing methods on both single video setting and continual setting, where spoof videos are interleaved with live ones to simulate spoofing attacks.
- Introduction to the Special Section on Deep Learning for Visual Surveillance2
The sheer amount of ground-based, aerial and mobile video surveillance data demands fittingly competent, accurate, effective techniques to extract useful cues and provide assistance for detection, prevention, and intervention tasks in traffic, safety, security, defense, forensic, health, biology, ethology, and retail space management applications.
- 2
- Sparse Coding on Cascaded Residuals2
This paper presents a two-pass multi-resolution cascade framework for dictionary learning and sparse coding, which is powerful in image coding, denoising, inpainting and artifact removal tasks outperforming the state-of-the-art techniques.
- Spatial encoding of visual words for image classification2
A unified representation that enhances BoVWs with explicit local and global structure models is presented that outperforms the conventional models and compares favorably to more sophisticated scene classification techniques.
- 2
- 2
- Enforcing Point-wise Priors on Binary Segmentation2
This work defines a graph Laplacian spectrum based cost function and embeds it into a minimization framework to extract image regions that are aligned on the object boundaries and also in accordance with the given point-wise priors.
- 2
- 2
- Rethinking RAG in Long Videos: What to Retrieve and How to Use It?1
This work introduces V-RAGBench, a benchmark of query, evidence chunk, answer$\rangle$ triplets over hour-scale videos, in which each answer depends on a unique evidence chunk, enabling decoupled evaluation of retrieval and generation, and CARVE, a training-free method that runs parallel retrievers across configurations and uses chunk-adaptive reranking to select a winning configuration for each chunk, which is then carried into generation.
- CLEAR: Closed-Loop Reinforcement Learning at Scale for End-to-End Autonomous Driving1
CLEAR, a system that enables closed-loop training using Reinforcement Learning (RL) at scale for E2E-AD and proposes to learn a novel residual waypoint policy around the waypoint prior from pretrained VLA policies, effectively harnessing the knowledge within.
- MAPLE: Latent Multi-Agent Play for End-to-End Autonomous Driving1
MAPLE achieves state-of-the-art driving performance on Bench2Drive and demonstrates scalable, closed-loop multi-agent play for robust E2E autonomous driving systems.
- Memory-Efficient Fine-Tuning Diffusion Transformers via Dynamic Patch Sampling and Block Skipping1
A memory-efficient fine-tuning framework called DiT-BlockSkip, integrating timestep-aware dynamic patch sampling and block skipping by precomputing residual features, which achieves competitive personalization performance qualitatively and quantitatively, while reducing memory usage substantially.
- Improving Optical Flow and Stereo Depth Estimation by Leveraging Uncertainty-Based Learning Difficulties1
The Occlusion Avoiding (OA) loss is proposed, designed to guide the network into cycle consistency-based confident regions, where feature matching is more reliable and effectively manage various types of challenging pixels and regions during training.
- 1
- 1
- 1
- 1
- 1
- Guidance Through Surrogate: Toward a Generic Diagnostic Attack1
The effect of label smoothing on AT is analyzed as one of the potential causes of gradient masking, leading to a novel attack dubbed guided projected gradient attack (G-PGA), which can be used as a diagnostic tool to reveal elusive robustness due to gradient masking in adversarial defenses.
- 1
- 1
- Learning to Generate Object Segment Proposals with Multi-modal Cues1
This paper presents a learning-based object segmentation proposal generation method for stereo images that makes use of learned deep features and designed geometric features to represent a region, as well as a learned similarity network to guide the grouping process.
- Video Object Segmentation via Super-trajectories1
A novel semi-supervised video segmentation approach based on an efficient video representation, called as "super-trajectory", which corresponds to a group of compact trajectories that exhibit consistent motion patterns, similar appearance and close spatiotemporal relationships.
- 1
- 1
- 1
- When VLAD Met Hilbert1
This paper addresses two fundamental limitations of VLAD: its requirement for the local descriptors to have vector form and its restriction to linear classifiers due to its high-dimensionality and proposes an approximate formulation that allows us to accelerate the coding process while still benefiting from the properties of kernel VL AD.
- 1
- Advanced Tracking Systems: Computational Approaches1
This book will focus on recent advances in computational approaches for detection and tracking of human body, road boundaries and lane markers as well as on recognition of human activities, drowsiness and distraction state.
- 1
- 1
- 1
- 1
- 1
- Compressed Domain Video Object1
The experimental results have shown that the proposed compressed domain video segmentation method provides the similar results as by using spatial domain process with much less computational complexity.
- Heli-Tele: Road Extraction from Helicopter Video1
A learning based road likelihood computation method that uses aerial imagery and fuses information from several weak features to align a color helicopter image onto a given street map to find address of buildings visible in the image.
- 1
- –
- –
- –
- SciFlow: Semantic Cross Interference for Self-Supervised Optical Flow Domain Generalization–
SciFlow is introduced, a simple yet effective, network-agnostic, training-based approach that leverages self-supervised learning to generalize motion estimation across synthetic and open-world domains and imposes semantic interference from open-world images onto synthetic images during training, which enables the network to adapt to the real-world domains.
- –
- –
- Learning Optical Flow Field via Neural Ordinary Differential Equation–
This paper introduces a novel approach for predicting the derivative of the flow using a continuous model, namely neural ordinary differential equations (ODE), and demonstrates that this approach achieves an impressive improvement over baseline and existing models, all while requiring only a single refinement step.
- –
- –
- –
- –
- Conditional Model Selection for Efficient Video Understanding–
This work trains a novel model-selector offline with model-affinity annotations that consolidate recognition quality and efficiency and incorporates the disparity between appearance and motion to estimate action background priors that enable efficient action localization without temporal annotations.
- –
- –
- –
- Selective Video Cutout using Global Pyramid Models and Local Uncertainty Propagation.–
A pyramid histogram based confidence map that incorporates structure information into appearance statistics that combines geodesic distance based dynamic models and employs an efficient measure of uncertainty propagation using local classifiers to determine the image regions where the object labels might be ambiguous.
- Light-Weight Single Image Super-Resolution via Pattern-wise Regression Function–
The goal is to generate high-resolution, artifact-less output images using reasonable computational resources and the evaluation result shows that the method is comparable to other state-of-the-art methods.
- –
- –
- Towards Practical Evaluation of Pedestrian Detectors–
Assessment and reason about the differences in detection performance of two state of the art detectors across changes in modality (visible or near infrared), evalution method (on cropped or whole images), the effect of different design choices (resizing features or images, smoothing or not).
- Using covariance improves computer detection and tracking of humans–
A novel object descriptor is developed, a bag of covariance matrices, to represent an image window, which is used to automatically detect and track any target object in video images.
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.