Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks419 from public data
- Generalized Intersection Over Union: A Metric and a Loss for Bounding Box Regression6,253
This paper introduces a generalized version of IoU ( GIoU) as a loss into the state-of-the art object detection frameworks, and shows a consistent improvement on their performance using both the standard, IoU based, and new, GIo U based, performance measures on popular object detection benchmarks.
- MonoSLAM: Real-Time Single Camera SLAM4,359
The first successful application of the SLAM methodology from mobile robotics to the "pure vision" domain of a single uncontrolled camera, achieving real time but drift-free performance inaccessible to structure from motion approaches is presented.
- Simultaneous Localization And Mapping: Present, Future, and the Robust-Perception Age.3,655
What is now the de-facto standard formulation for SLAM is presented, covering a broad set of topics including robustness and scalability in long-term mapping, metric and semantic representations for mapping, theoretical performance guarantees, active SLAM and exploration, and other new frontiers.
- RefineNet: Multi-path Refinement Networks for High-Resolution Semantic Segmentation3,218
RefineNet is presented, a generic multi-path refinement network that explicitly exploits all the information available along the down-sampling process to enable high-resolution prediction using long-range residual connections and introduces chained residual pooling, which captures rich background context in an efficient manner.
- Past, present, and future of simultaneous localization and mapping: Toward the robust-perception age3,128
- Vision-and-Language Navigation: Interpreting Visually-Grounded Navigation Instructions in Real Environments2,014
This work provides the first benchmark dataset for visually-grounded natural language navigation in real buildings - the Room-to-Room (R2R) dataset and presents the Matter-port3D Simulator - a large-scale reinforcement learning environment based on real imagery.
- Unsupervised CNN for Single View Depth Estimation: Geometry to the Rescue1,670
This work proposes a unsupervised framework to learn a deep convolutional neural network for single view depth prediction, without requiring a pre-training stage or annotated ground-truth depths, and shows that this network trained on less than half of the KITTI dataset gives comparable performance to that of the state-of-the-art supervised methods for singleView depth estimation.
- 1,373
- Learning Depth from Single Monocular Images Using Deep Convolutional Neural Fields1,303
A deep convolutional neural field model for estimating depths from single monocular images, aiming to jointly explore the capacity of deep CNN and continuous CRF is presented, and a deep structured learning scheme which learns the unary and pairwise potentials of continuousCRF in a unified deep CNN framework is proposed.
- 1,096
- Efficient Piecewise Training of Deep Structured Models for Semantic Segmentation941
This work shows how to improve semantic segmentation through the use of contextual information, specifically, ' patch-patch' context between image regions, and 'patch-background' context, and formulate Conditional Random Fields with CNN-based pairwise potential functions to capture semantic correlations between neighboring patches.
- 860
- 781
- 732
- Unsupervised Learning of Monocular Depth Estimation and Visual Odometry with Deep Feature Reconstruction703
The use of stereo sequences for learning depth and visual odometry enables the use of both spatial and temporal photometric warp error, and constrains the scene depth and camera motion to be in a common, real-world scale.
- Online Multi-Target Tracking Using Recurrent Neural Networks549
This work proposes for the first time, an end-to-end learning approach for online multi-target tracking based on recurrent neural networks (RNNs) and shows promising results obtained at ~300 Hz on a standard CPU.
- 544
- SayPlan: Grounding Large Language Models using 3D Scene Graphs for Scalable Robot Task Planning535
SayPlan is introduced, a scalable approach to LLM-based, large-scale task planning for robotics using 3D scene graph (3DSG) representations that exploit the hierarchical nature of 3DSGs to allow LLMs to conduct a 'semantic search' for task-relevant subgraphs from a smaller, collapsed representation of the full graph.
- 503
- Articulated Body Motion Capture by Stochastic Search482
A modified particle filter is developed which is shown to be effective at searching the high-dimensional configuration spaces encountered in visual tracking of articulated body motion and to be capable of recovering full articulated bodymotion efficiently.
- 467
- From Motion Blur to Motion Flow: A Deep Learning Solution for Removing Heterogeneous Motion Blur464
This work directly estimates the motion flow from the blurred image through a fully-convolutional deep neural network (FCN) and recovers the unblurred image from the estimated motion flow and is the first universal end-to-end mapping from the blur image to the dense motion flow.
- MOTChallenge: A Benchmark for Single-Camera Multiple Target Tracking418
This paper collects the first three releases of MOTChallenge and provides a categorization of state-of-the-art trackers and a broad error analysis, to help newcomers understand the related work and research trends in the MOT community, and hopefully shed some light into potential future research directions.
- 394
- Multi-modal Cycle-Consistent Generalized Zero-Shot Learning389
This paper proposes the use of a constraint based on a new regularization for the GAN training that forces the generated visual features to reconstruct their original semantic features in a multi-modal cycle-consistent manner.
- AffordanceNet: An End-to-End Deep Learning Approach for Object Affordance Detection382
The experimental results on the public datasets show that the AffordanceNet outperforms recent state-of-the-art methods by a fair margin, while its end-to-end architecture allows the inference at the speed of 150ms per image.
- Smart Mining for Deep Metric Learning377
This paper proposes a novel deep metric learning method that combines the triplet model and the global structure of the embedding space and relies on a smart mining procedure that produces effective training samples for a low computational cost.
- 371
- Deep learning features at scale for visual place recognition366
This paper trains, at large scale, two CNN architectures for the specific place recognition task and employs a multi-scale feature encoding method to generate condition- and viewpoint-invariant features.
- Learning Local Image Descriptors with Deep Siamese and Triplet Convolutional Networks by Minimizing Global Loss Functions322
A combination of the triplet and global losses produces the best embedding in the field, using this triplet network, and it is demonstrated that the use of the central-surround siamese network trained with the global loss producing the best result of the field on the UBC dataset.
- Attention-Guided Network for Ghost-Free High Dynamic Range Imaging313
The proposed AHDRNet is a non-flow-based method, which can also avoid the artifacts generated by optical-flow estimation error, and can achieve state-of-the-art quantitative and qualitative results.
- Mapping Large Loops with a Single Hand-Held Camera305
This paper presents a method for Simultaneous Localization and Mapping (SLAM) relying on a monocular camera as the only sensor which is able to build outdoor, closedloop maps much larger than previously achieved with such input.
- Real-Time Monocular SLAM with Straight Lines297
This work describes how straight lines can be added to a monocular Extended Kalman Filter Simultaneous Mapping and Localisation (EKF SLAM) system in a manner that is both fast and which integrates easily with point features.
- 285
- 278
- Towards Effective Low-Bitwidth Convolutional Neural Networks266
This paper tackles the problem of training a deep convolutional neural network with both low-precision weights and low-bitwidth activations by proposing a two-stage optimization strategy to progressively find good local minima and adopting a novel learning scheme to jointly train a full- Precision model alongside the low-Precision one.
- A plane measuring device263
An uncertainty analysis which includes both the errors in image localization and the uncertainty in the imaging transformation is developed, and the distribution of correspondences can be chosen to achieve a particular bound on the uncertainty.
- Deep Regression Tracking with Shrinkage Loss258
This work proposes a novel shrinkage loss to penalize the importance of easy training data and applies residual connections to fuse multiple convolutional layers as well as their output response maps to balance training data.
- RSLAM: A System for Large-Scale Mapping in Constant-Time Using Stereo236
A relative simultaneous localisation and mapping system (RSLAM) for the constant-time estimation of structure and motion using a binocular stereo camera system as the sole sensor and a topo-metric representation in terms of a sequence of relative locations is described.
- Meaningful maps – Object-oriented semantic mapping231
This paper simultaneously build geometric point cloud models of previously unseen instances of known object classes and create a map that contains these object models as central entities that leverages sparse, feature-based RGB-D SLAM, image-based deep-learning object detection and 3D unsupervised segmentation.
- Meaningful maps with object-oriented semantic mapping231
This paper simultaneously build geometric point cloud models of previously unseen instances of known object classes and create a map that contains these object models as central entities that leverages sparse, feature-based RGB-D SLAM, image-based deep-learning object detection and 3D unsupervised segmentation.
- Unsupervised Scale-Consistent Depth Learning from Video226
A monocular depth estimation method SC-Depth, which requires only unlabelled videos for training and enables the scale-consistent prediction at inference time and the proposed hybrid Pseudo-RGBD SLAM shows compelling results in KITTI, and it generalizes well to the KAIST dataset without additional training.
- Robust Real-Time Visual Tracking Using Pixel-Wise Posteriors224
A probabilistic framework for robust, real-time, visual tracking of previously unseen objects from a moving camera is derived using a bag-of-pixels representation and the key to the success is the use of pixel-wise posteriors, as opposed to likelihoods.
- fastHOG – a real-time GPU implementation of HOG219
A parallel implementation of the histogram of oriented gradients algorithm for object detection, using the GPU and the NVIDIA CUDA framework, which allows for real-time performance of the full HOG algorithm for the first time in the literature.
- Visual Odometry Revisited: What Should Be Learnt?207
This work revisit the basics of VO and explore the right way for integrating deep learning with epipolar geometry and Perspective-n-Point method and design a simple but robust frame-to-frame VO algorithm (DF-VO) which outperforms pure deep learning-based and geometry-based methods.
- 203
- PWP3D: Real-Time Segmentation and Tracking of 3D Objects200
These experiments demonstrate the benefit of using pixel-wise posteriors rather than likelihoods, and showcase the qualities, such as robustness to occlusions and motion blur (and also some failure modes), of the tracker.
- High Five: Recognising human interactions in TV shows198
A per-person descriptor that uses attention (head orientation) and the local spatial and temporal context in a neighbourhood of each detected person and a structured SVM that combines head orientation and the relative location of people in a frame to improve upon the initial classification obtained with the descriptor.
- A general method for human activity recognition in video197
A system for human behaviour recognition in video sequences that combines Bayesian networks and belief propagation, non-parametric sampling from a previously learned database of actions, and Hidden Markov Models which encode scene rules are used to smooth sequences of actions.
- 191
- Adaptive relative bundle adjustment188
This paper derives a new relative bundle adjustment, which instead of optimizing in a single Euclidean space, works in a metric-space defined by a connected Riemannian manifold, and shows experimentally that it is possible to solve for the full ML solution incrementally in constant time – even at loop closure.
- 183
- Vast-scale Outdoor Navigation Using Adaptive Relative Bundle Adjustment182
A new relative bundle adjustment is derived which, instead of optimizing in a single Euclidean space, works in a metric space defined by a manifold, and it is shown experimentally that it is possible to solve for the full maximum-likelihood solution incrementally in constant time, even at loop closure.
- 179
- Self-Calibration of Rotating and Zooming Cameras179
The theory and practice of self-calibration of cameras which are fixed in location and may freely rotate while changing their internal parameters by zooming is described and some near-ambiguities that arise under rotational motions are identified.
- Fast Neural Architecture Search of Compact Semantic Segmentation Models via Auxiliary Cells176
This work focuses on dense per-pixel tasks, in particular, semantic image segmentation using fully convolutional networks, and relies on a progressive strategy that terminates non-promising architectures from being further trained, and on Polyak averaging coupled with knowledge distillation to speed-up the convergence.
- 173
- 169
- Towards Context-Aware Interaction Recognition for Visual Relationship Detection168
This paper proposes an alternative, context-aware interaction recognition framework, and explicitly construct an interaction classifier which combines the context, and the interaction, to avoid the limitation of both approaches.
- Locally Planar Patch Features for Real-Time Structure from Motion167
This work shows that long-term tracking is improved by treating salient feature patches as observations of locally planar regions on 3D world surfaces on SLAM framework, enabling pre-warping of templates for matching.
- Structured Binary Neural Networks for Accurate Image Classification and Semantic Segmentation166
This paper proposes a ``network decomposition'' strategy, named Group-Net, in which each full-precision group can be effectively reconstructed by aggregating a set of homogeneous binary branches, and shows strong generalization to other tasks.
- Real-Time Joint Semantic Segmentation and Depth Estimation Using Asymmetric Annotations166
This work adapts a recently proposed real-time semantic segmentation network, making changes to further reduce the number of floating point operations, and incorporates the raw predictions of the network into the SemanticFusion framework for dense 3D semantic reconstruction of the scene.
- 164
- 160
- On the comparison of uncertainty criteria for active SLAM155
It is shown that contrary to what has been previously reported, the D-optimality criterion is indeed capable of giving fruitful information as a metric for the uncertainty of a robot performing SLAM.
- Semantics for Robotic Mapping, Perception and Interaction: A Survey154
A taxonomy for semantics research in or relevant to robotics is established, split into four broad categories of activity, in which semantics are extracted, used, or both, and survey dozens of major topics including fundamentals from the computer vision field and key robotics research areas utilizing semantics.
- Parallel Attention: A Unified Framework for Visual Object Discovery Through Dialogs and Queries153
A unified framework, the ParalleL AttentioN (PLAN) network, to discover the object in an image that is being referred to in variable length natural expression descriptions, from short phrases query to long multi-round dialogs is proposed.
- 151
- Cartman: The Low-Cost Cartesian Manipulator that Won the Amazon Robotics Challenge149
The design of the custom-built, cost-effective, Cartesian robot system Cartman, which won first place in the competition finals by stowing 14 and picking all 9 items in 27 minutes, scoring a total of 272 points, is presented.
- Are You Talking to Me? Reasoned Visual Dialog Generation Through Adversarial Learning148
A novel approach that combines Reinforcement Learning and Generative Adversarial Networks (GANS) to generate more human-like responses to questions to overcome the relative paucity of training data, and the tendency of the typical MLE-based approach to generate overly terse answers.
- Navigating, Recognizing and Describing Urban Spaces With Vision and Lasers148
A body of work aimed at extending the reach of mobile navigation and mapping is described, showing how running topological and metric mapping and pose estimation processes concurrently, using vision and laser ranging, has produced a full six-degree-of-freedom outdoor navigation system.
- Fast Training of Triplet-Based Deep Binary Embedding Networks138
This paper proposes to formulate high-order binary codes learning as a multi-label classification problem by explicitly separating learning into two interleaved stages and proposes to map the original image to compact binary codes via carefully designed deep convolutional neural networks and the hashing function fitting can be solved by training binary CNN classifiers.
- A Constant-Time Efficient Stereo SLAM System136
This article introduces a simple but novel representation of the environment in terms of a sequence of relative locations, and demonstrates precise local mapping and easy navigation using the relative map, and importantly shows that this can be done without requiring a global minimisation after loop closure.
- 135
- 135
- SC-DepthV3: Robust Self-Supervised Monocular Depth Estimation for Dynamic Scenes134
This work introduces an external pretrained monocular depth estimation model for generating single-image depth prior, namely pseudo-depth, based on which it proposes novel losses to boost self-supervised training and demonstrates the significantly superior performance of the method over previous methods on six challenging datasets.
- Exploring Context with Deep Structured Models for Semantic Segmentation134
This work formulate deep structured models by combining CNNs and Conditional Random Fields for learning the patch-patch context between image regions, and formulate CNN-based pairwise potential functions to capture semantic correlations between neighboring patches.
- GMS: Grid-Based Motion Statistics for Fast, Ultra-robust Feature Correspondence133
GMS is proposed, which incorporates the smoothness constraint into a statistic framework for separation and uses a grid-based implementation for fast calculation and integrates into the well-known ORB-SLAM system for monocular initialization, resulting in a significant improvement.
- 128
- gSLIC: a real-time implementation of SLIC superpixel segmentation127
A parallel implementation of the Simple Linear Iterative Clustering (SLIC) superpixel segmentation that uses GPU and the NVIDIA CUDA framework and achieves speedups of 10x∼20x from the standard sequential implementation.
- 123
- Real-time Image Smoothing via Iterative Least Squares121
A new global optimization based method, named iterative least squares (ILS), for efficient edge-preserving image smoothing, which can produce high-quality results but at a much lower computational cost.
- Guiding Visual Surveillance by Tracking Human Attention121
A system for automatic tracking and detection of individual heads to infer their gaze direction is developed and both static and temporally varying maps of areas where people look are able to identify interesting regions.
- LongReMix: Robust learning with high confidence samples in a noisy label environment119
This paper empirically validate two hypotheses and proposes the new 2-stage noisy-label training algorithm LongReMix, which generalises better than competing approaches, particularly in high label noise problems and achieves state-of-the-art performance in most datasets.
- 113
- Visual Question Answering with Memory-Augmented Networks112
It is shown that memory-augmented neural networks are able to maintain a relatively long-term memory of scarce training exemplars, which is important for visual question answering due to the heavy-tailed distribution of answers in a general VQA setting.
- Deep Reinforcement Learning for Active Breast Lesion Detection from DCE-MRI112
The deep Q-network approach is extended, previously demonstrated on simpler problems such as anatomical landmark detection, in order to detect lesions that have a significant variation in shape, appearance, location and size.
- 111
- RGBD Based Dimensional Decomposition Residual Network for 3D Semantic Scene Completion107
A light-weight Dimensional Decomposition Residual network (DDR) for 3D dense prediction tasks is introduced and the proposed multi-scale fusion mechanism for depth and color image can improve the completion and segmentation accuracy simultaneously.
- 107
- The active recovery of 3D motion trajectories and their use in prediction107
It is demonstrated that prediction from the 3D planar description gives better tracking performance than prediction based on a filtered description of observer-based 2D motion trajectories.
- Simultaneous Localisation and Mapping in Dynamic Environments (SLAMIDE) with Reversible Data Associa106
It is shown empirically that if dynamic objects are present the method can include them in a single framework and hence maintain a consistent estimate and the estimator remains consistent when data association is difficult, for instance in the presence of clutter.
- 104
- Multi-modal Auto-Encoders as Joint Estimators for Robotics Scene Understanding103
It is shown that suitably designed Multi-modal Auto-Encoders can solve the depth estimation and the semantic segmentation problems simultaneously, in the partial or even complete absence of some of the input modalities.
- Goal-directed video metrology103
It is shown, via a specific example — plotting the position of a goal-bound soccer ball — that accurate measurements can be obtained, and that both qualitative and quantitative questions about the data can be answered.
- 99
- A modular head/eye platform for real-time reactive vision99
This paper describes the design, implementation and testing of a high speed controlled stereo “head/eye” platform which facilitates the rapid redirection of gaze in response to visual input.
- 98
- Estimating Gaze Direction from Low-Resolution Faces in Video98
A new method for automatically estimating where a person is looking in images where the head is typically in the range 20 to 40 pixels high, using a feature vector based on skin detection to estimate the orientation of the head, relative to the camera.
- Active tracking of foveated feature clusters using affine structure98
The method makes use of a real-time implementation of a corner detector and tracker and reconstructs the image position of the desired fixation point from a cluster of corners detected on the object using the affine structure available from two or three views.
- Socially and Contextually Aware Human Motion and Pose Forecasting96
A novel framework to tackle both tasks of human motion (or trajectory) and body skeleton pose forecasting in a unified end-to-end pipeline is proposed and achieves a superior performance compared to several baselines on two social datasets.
- 96
- Dense 3D Face Correspondence92
An algorithm that automatically establishes dense correspondences between a large number of 3D faces and iterates between rigid alignment of an unseen face followed by regularized morphing of the deformable model.
- 92
- Attend in Groups: A Weakly-Supervised Deep Learning Framework for Learning from Web Data91
This work proposes an end-to-end weakly-supervised deep learning framework which is robust to the label noise in Web images and relies on two unified strategies, random grouping and attention, to effectively reduce the negative impact of noisy web image annotations.
- Joint Learning of Social Groups, Individuals Action and Sub-group Activities in Videos90
This paper proposes an end-to-end trainable framework for the social task, and sets the state-of-the-art results on two widely adopted benchmarks for the traditional group activity recognition task.
- 90
- Auto-Rectify Network for Unsupervised Indoor Depth Estimation89
This work establishes that the complex ego-motions exhibited in handheld settings are a critical obstacle for learning depth, and proposes an Auto-Rectify Network with novel loss functions, which can automatically learn to rectify images during training.
- 87
- 86
- 85
- Self-Calibration of a Rotating Camera with Varying Intrinsic Parameters85
This work presents a method for self-calibration of a camera which is free to rotate and change its intrinsic parameters, but which cannot translate and gives experimental results using real image sequences for which ground truth data was available.
- 84
- 84
- Real-Time Monocular Object-Model Aware Sparse SLAM83
This work incorporates a real- time deep-learned object detector to the monocular SLAM framework for representing generic objects as quadrics that permit detections to be seamlessly integrated while allowing the real-time performance.
- Scalable Place Recognition Under Appearance Change for Autonomous Driving83
This work proposes a novel place recognition technique that can be efficiently retrained and compressed, such that the recognition of new queries can exploit all available data (including recent changes) without suffering from visible growth in computational cost.
- FroDO: From Detections to 3D Objects81
FroDO is a method for accurate 3D reconstruction of object instances from RGB video that infers their location, pose and shape in a coarse to fine manner to embed object shapes in a novel learnt shape space that allows seamless switching between sparse point cloud and dense DeepSDF decoding.
- 81
- Driving saccade to pursuit using image motion81
A methodology for, and real-time demonstrations of, the use of motion detection and segmentation processes to initiate “capture saccades” towards a moving object, and demonstrates in repeated trials that the transition from saccadic motion to tracking is more likely to succeed using position and velocity control, than when using position alone.
- Residual Pattern Learning for Pixel-wise Out-of-Distribution Detection in Semantic Segmentation80
A new residual pattern learning (RPL) module that assists the segmentation model to detect OoD pixels with minimal deterioration to inlier segmentation accuracy; and a novel context-robust contrastive learning (CoroCL) that enforces RPL to robustly detect OoD pixels in various contexts.
- Depth Based Semantic Scene Completion With Position Importance Aware Loss79
PALNet is proposed, a novel hybrid network for SSC based on single depth using a two-stream network to extract both 2D and 3D features from multi-stages using fine-grained depth information to efficiently capture the context, as well as the geometric cues of the scene.
- Constraint-Aware Zero-Shot Vision-Language Navigation in Continuous Environments77
A Constraint-Aware Navigator (CA-Nav), which reframes zero-shot VLN-CE as a sequential, constraint-aware sub-instruction completion process, and achieves the state-of-the-art performance on two VLN-CE benchmarks, surpassing the previous best method by 12% and 13% in Success Rate on the validation unseen splits of R2R-CE and RxR-CE, respectively.
- TRiPOD: Human Trajectory and Pose Dynamics Forecasting in the Wild77
A novel TRajectory and POse Dynamics (nicknamed TRiPOD) method based on graph attentional networks to model the human-human and human-object interactions both in the input space and the output space (decoded future output).
- Training Quantized Neural Networks With a Full-Precision Auxiliary Module76
The proposed method achieves near lossless performance to the full-precision model by using a 4-bit detector, which is of great practical value, and evaluates the proposed method on image classification and object detection over various quantization approaches and show consistent performance increase.
- A Theoretically Sound Upper Bound on the Triplet Loss for Improving the Efficiency of Deep Distance Metric Learning76
A training algorithm that no longer requires the centroid optimization step is proposed, which means that this approach is the first in the field with a guaranteed linear run-time complexity.
- 75
- 73
- 73
- 72
- Looking Beyond Two Frames: End-to-End Multi-Object Tracking Using Spatial and Temporal Transformers71
MO3TR is presented: a truly end-to-end Transformer-based online multi-object tracking (MOT) framework that learns to handle occlusions, track initiation and termination without the need for an explicit data association module or any heuristics.
- 71
- Recursive affine structure and motion from image sequences71
This paper presents a new algorithm for structure from motion from an arbitrary number of tracked features over an arbitrarily number of images, which possesses several advantages over previous formulations.
- JRDB-Act: A Large-scale Dataset for Spatio-temporal Action, Social Group and Activity Detection70
This paper introduces JRDB-Act, an extension of the existing JRDB, which is captured by a social mobile manipulator and reflects a real distribution of human daily-life actions in a university campus environment and develops an end-to-end trainable pipeline to learn and infer these tasks, i.e. individual action and social group detection.
- 68
- Projective calibration of a laser-stripe range finder68
The calibration of a laser triangulation range system mounted on a robot vehicle is discussed, and the Oxford/NEL range finder, a sensor mounted on the Oxford AGV for use in object recognition and acquisition tasks, is described.
- Structure Aware SLAM Using Quadrics and Planes65
This work proposes a method for representing generic objects as quadrics which allows object detections to be seamlessly integrated in a SLAM framework and can easily incorporate Manhattan and object affordance constraints, greatly improving camera localization and leading to semantically meaningful maps.
- Simultaneous Monocular 2D Segmentation, 3D Pose Recovery and 3D Reconstruction65
The solution is to learn nonlinear and probabilistic low dimensional latent spaces, using the Gaussian Process Latent Variable Models dimensionality reduction technique, which acts as class or activity constraints to a simultaneous and variational segmentation --- recovery --- reconstruction process.
- 65
- Data-Driven Approximations to NP-Hard Problems64
This paper proposes a simple, yet effective technique for improving the initial (weak) training set by incorporating the objective cost into the training procedure, and demonstrates the functionality of the approach on three exemplar applications: marginal distributions of a joint matching space, feature point matching and the travelling salesman problem.
- Colour Invariant Head Pose Classification in Low Resolution Video64
Results from video sequences demonstrate that an improved posterior estimation using learnt colour distributions reduces classification error and provides accurate pose information in images where the head occupies as little as 10 pixels square.
- Duality, rigidity and planar parallax63
The geometry of two views of seven points and three views of six points are investigated, it is proved that the two are dual, and that the fundamental geometric constraints in each case are encapsulated by a planar homology.
- NeuRoRA: Neural Robust Rotation Averaging62
This work aims to build a neural network that learns the noise patterns from the data and predict/regress the model parameters from the noisy relative orientations and replaces robust optimization methods by a graph-based network.
- Markerless Motion Capture of Complex Full-Body Movement for Character Animation62
The approach to searching through the high-dimensional model configuration space is an algorithm called annealed particle filtering which finds the best fit to image data via multiple-layer propagation of a stochastic particle set.
- 60
- Training Medical Image Analysis Systems like Radiologists60
The use of meta-training that models a classifier based on a series of tasks is explored, which is shown to be the best in the field for that application, compared to state of art baseline approaches: DenseNet, multiple instance learning and multi-task learning.
- Saccade and pursuit on an active head/eye platform60
A realtime active surveillance vision system which detects moving objects in an everyday environment, directs the gaze of a headkye platform towards the objects and subsequently pursues them smoothly, and can continue over extended periods.
- SG-VAE: Scene Grammar Variational Autoencoder to Generate New Indoor Scenes59
This work proposes a neural network to learn a generative model for sampling consistent indoor scene layouts that learns the co-occurrences, and appearance parameters such as shape and pose, for different objects categories through a grammar-based auto-encoder, resulting in a compact and accurate representation for scene layouts.
- A hybrid probabilistic model for camera relocalization59
The proposed system leverages the discriminative deep image representation from a convolutional neural networks, and uses Gaussian Process regressors to generate the probability distribution of the six degree of freedom camera pose in an end-to-end fashion, resulting in a network that can generate uncertainties over its inferences with no need to sample many times.
- 59
- 59
- 59
- RoboHop: Segment-based Topological Map Representation for Open-World Visual Navigation58
A novel topological representation of an environment based on , which are semantically meaningful and open-vocabulary queryable, conferring several advantages over previous works based on pixel-level features.
- 58
- 58
- DeepSetNet: Predicting Sets with Deep Neural Networks57
This paper defines a likelihood for a set distribution and learns its parameters using a deep neural network and derives a loss for predicting a discrete distribution corresponding to set cardinality.
- 3D hand tracking for human computer interaction56
This work proposes a real-time model-based 3D hand tracker that combines image regions and the signal from an off-the-shelf 3-axis accelerometer placed on the user's hand to distinguish between a number of activity dependent articulated hand configurations.
- Effective Training of Convolutional Neural Networks With Low-Bitwidth Weights and Activations55
This paper tackles the problem of training a deep convolutional neural network of both low-bitwidth weights and activations and proposes three practical approaches, including progressive quantization; stochastic precision; and joint knowledge distillation to improve the network training.
- 55
- 54
- 54
- ActiveRMAP: Radiance Field for Active Mapping And Planning53
This paper presents an RGB-only active vision framework using radiance field representation for active 3D reconstruction and planning in an online manner and suggests that the proposed method achieves competitive results compared to other offline methods and outperforms active reconstruction methods using NeRFs.
- 53
- 53
- Self-supervised Mean Teacher for Semi-supervised Chest X-Ray Classification52
Self-supervised Mean Teacher for Semi- Supervised (S$2$MTS$^2$) learning that combines self-super supervised mean-teacher pre-training with semi-supervisory fine-tuning is proposed that outperforms the previous SOTA semi- supervised learning methods by a large margin.
- Semantic Segmentation from Limited Training Data52
This work presents a deep metric learning approach that works in three separate steps: semantic-agnostic boundary detection, patch classification and pixel-wise voting and a fully-supervised semantic segmentation approach with efficient dataset collection for robotic perception in cluttered scenes.
- Synchronizing Image Sequences of Non-Rigid Objects52
A simple but effective method for automatically recovering the sub-frame temporal offset between image sequences taken using unsynchronized cameras is presented, and the affine structure of a non-rigid motion is obtained.
- Dual-Attention-Guided Network for Ghost-Free High Dynamic Range Imaging51
This work proposes a novel dual-attention-guided end-to-end deep neural network, called DAHDRNet, which produces high-quality ghost-free HDR images and employs a hybrid loss function, which consists of a perceptual loss, a total variation loss, and a content loss to recover photo-realistic images.
- 51
- Motion Anything: Any to Motion Generation50
Motion Anything is proposed, a multimodal motion generation framework that introduces an Attention-based Mask Modeling approach, enabling fine-grained spatial and temporal control over key frames and actions, and introduces Text-Music-Dance (TMD), a new motion dataset consisting of 2,153 pairs of text, music, and dance.
- EvidentialMix: Learning with Combined Open-set and Closed-set Noisy Labels50
A new variant of the noisy label problem that combines the open-set and closed-set noisy labels is studied, and a novel algorithm is proposed, called EvidentialMix, that addresses this problem and is compared with the state-of-the-art methods for both closed- set and open- set noise on the proposed benchmark.
- A Dynamic Programming Approach to Reconstructing Building Interiors50
A dynamic programming solution to the reconstruction problem for "indoor" Manhattan worlds (a sub-class of Manhattan worlds), which deterministically finds the global optimum and exhibits computational complexity linear in both model complexity and image size.
- When LLMs step into the 3D World: A Survey and Meta-Analysis of 3D Tasks via Multi-modal Large Language Models49
This survey provides a comprehensive overview of the methodologies enabling LLMs to process, understand, and generate 3D data, and examines their integration with LLMs for tasks such as 3D scene understanding, captioning, question-answering, and dialogue.
- SceneCut: Joint Geometric and Object Segmentation for Indoor Scenes49
sceneCut's joint reasoning over scene semantics and geometry allows a robot to detect and segment object instances in complex scenes where modern deep learning-based methods either fail to separate object instances, or fail to detect objects that were not seen during training.
- HCVRD: A Benchmark for Large-Scale Human-Centered Visual Relationship Detection48
A webly-supervised approach to these problems is proposed and it is demonstrated that the proposed model provides a strong baseline on the authors' HCVRD dataset.
- 48
- Design of stereo heads48
The design considerations behind the construction of a steerable stereo head/eye platform, the mechanical design and that of the gaze controller and vision systems are outlined, and it is shown how the kinematics of the platform can be used to relate image measurements with the scene motion in a static frame.
- 47
- Visual SLAM: Why Bundle Adjust?47
This work proposes a different SLAM optimisation core: instead of bundle adjustment, it conducts rotation averaging to incrementally optimise only camera orientations and obviates the need to estimate and maintain the positions and 3D map at keyframe rate, which enables simpler SLAM systems.
- 47
- The Pathfinder image-guided surgical robot46
The Pathfinder image-guided surgical robot that has been designed to replace the stereotactic frame in neurosurgery is described, and the calibration stages employed in order to achieve submillimetre positioning accuracy of a tool tip are detailed.
- 45
- 44
- 44
- Video synchronization from human motion using rank constraints43
An efficient method is presented that recovers potential frame correspondences, estimates possible synchronization parameters via the Hough transform and refines these parameters using non-linear optimization methods in order to recover synchronization to sub-frame accuracy, even for sequences of unknown and different frame rates.
- On New View Synthesis Using Multiview Stereo43
It is shown that application of modern multiview stereo techniques to the newview synthesis (NVS) problem introduces a number of non-trivial complexities, and a synthesis of the two approaches which yields good results on difficult image sequences is addressed.
- Scalable Deep k-Subspace Clustering41
This paper introduces a method that simultaneously learns an embedding space along subspaces within a deep neural network to minimize a notion of reconstruction error, thus addressing the problem of subspace clustering in an end-to-end learning paradigm.
- Providing synthetic views for teleoperation using visual pose tracking in multiple cameras40
This paper describes a visual tool for teleoperative experimentation involving remote manipulation and contact tasks that recovers in real time the pose of moving polyhedral objects, and presents a synthetic view of the scene to the operator of a teleoperated robot using any chosen viewpoint and viewing direction.
- 39
- 38
- 38
- Efficient Dense Point Cloud Object Reconstruction Using Deformation Vector Fields37
A novel approach that addresses the limitation of existing CNN-based methods for single-view 3D object reconstruction by replacing masks with “deformation-fields”, which encourages the 3D points back-projected from a particular view to be consistent across views.
- Addressing Challenging Place Recognition Tasks Using Generative Adversarial Networks37
This work addresses place recognition as a domain translation task using a pair of coupled Generative Adversarial Networks to show that it is possible to generate the appearance of one domain from another without requiring image-to-image correspondences across the domains.
- 37
- Bayesian Semantic Instance Segmentation in Open Set World35
This paper presents a novel open-set semantic instance segmentation approach capable of segmenting all known and unknown object classes in images, based on the output of an object detector trained on known object classes.
- ODAM: Object Detection, Association, and Mapping using Posed RGB Video34
ODAM, a system for 3D Object Detection, Association, and Mapping using posed RGB videos, relies on a deep learning front-end to detect 3D objects from a given RGB frame and associate them to a global object-based map using a graph neural network (GNN).
- 34
- 34
- Real-Time Tracking of Single and Multiple Objects from Depth-Colour Imagery Using 3D Signed Distance Functions34
This work presents first a method suited to tracking a single rigid 3D object, and then generalise this to multiple objects by combining distance functions into a shape union in the frame of the camera, leading to robust real-time tracking without recourse to bolt-on or ad-hoc collision detection.
- Fields of Experts for Image-based Rendering34
A new optimization algorithm from statistical physics is introduced to computer vision for the first time which reaches better minima than the ICM and simulated annealing algorithms to which such large-clique problems have previously been restricted.
- How Trustworthy are Performance Evaluations for Basic Vision Tasks?33
A notion of trustworthiness for performance criteria is suggested, which requires robustness to parameters for reliability, contextual meaningfulness in sanity tests, and consistency with mathematical requirements such as the metric properties.
- Unsupervised Depth Learning in Challenging Indoor Video: Weak Rectification to Rescue.33
This work establishes that the degenerate camera motions exhibited in handheld settings are a critical obstacle for unsupervised depth learning and proposes a novel data pre-processing method for effective training, i.e., search for image pairs with modest translation and remove their rotation via the proposed weak image rectification.
- Self-supervised Learning for Single View Depth and Surface Normal Estimation33
It is shown that a simple depth-normal consistency as a soft-constraint on the predictions is sufficient and effective for training both these networks simultaneously and outperforms the traditional self-supervised depth prediction network by a large margin on the KITTI benchmark.
- Seeing Behind Things: Extending Semantic Segmentation to Occluded Regions33
This work proposes a modification of the standard cross-entropy loss to cope with the settings of semantic segmentation, and demonstrates that a CNN trained by minimizing the proposed loss is able to predict semantic categories for visible and occluded object parts without requiring to increase the network size.
- Pre and post-hoc diagnosis and interpretation of malignancy from breast DCE-MRI33
A new method for breast cancer screening from DCE-MRI based on a post-hoc approach that is trained using weakly annotated data, which is more accurate for diagnosing the whole volume per patient and localizes the malignant lesions that led to such diagnosis.
- Just-in-Time Reconstruction: Inpainting Sparse Maps Using Single View Depth Predictors as Priors33
This work adopts a fairly standard approach to data fusion, to produce a fused depth map by performing inference over a novel fully-connected Conditional Random Field (CRF) which is parameterized by the input depth maps and their pixel-wise confidence weights.
- 33
- A Unified Energy Minimization Framework for Model Fitting in Depth33
This paper presents a unified energy minimization framework for model fitting and pose recovery problems in depth cameras, which takes the advantage of the gradient of 3D level-set embedding function and can be efficiently solved by gradients-based optimization methods.
- COMBINING TRAFFIC SIGN DETECTION WITH 3D TRACKING TOWARDS BETTER DRIVER ASSISTANCE33
A realtime version of driver assistance systems that integrates single view detection with region-based 3D tracking of traffic signs and obtains 3D pose information that is used to establish the relevance of the traffic sign to the driver.
- 3D Trajectories from a Single Viewpoint using Shadows33
This work considers the problem of obtaining the 3D trajectory of a ball from a sequence of images taken with a camera which is possibly rotating and zooming (but not translating), and develops techniques to compute the component of image motion of the ball due to camera rotation and zoom using optic flow.
- Hier-SLAM: Scaling-Up Semantics in SLAM with a Hierarchically Categorical Gaussian Splatting32
Hier-SLAM is proposed, a semantic 3D Gaussian Splatting SLAM method featuring a novel hierarchical categorical representation that encodes semantic information in a compact form into 3D Gaussian Splatting, leveraging the capabilities of large language models (LLMs).
- 29
- Automatic Reasoning about Causal Events in Surveillance Video29
A new method for explaining causal interactions among people in video using the low-level vision techniques of spatiotemporal action recognition and gaze-direction approximation, which is a general sensing strategy for a person agent in a variety of application domains.
- Cognitive visual tracking and camera control29
An efficient solution to extract, in real-time, high-level information from an observed scene, and generate the most appropriate commands for a set of pan-tilt-zoom (PTZ) cameras in a surveillance scenario is presented.
- Approximate Fisher Information Matrix to Characterize the Training of Deep Neural Networks28
A novel methodology for characterizing the performance of deep learning networks (ResNets and DenseNet) with respect to training convergence and generalization as a function of mini-batch size and learning rate for image classification based on novel measurements derived from the eigenvalues of the approximate Fisher information matrix.
- Weakly Supervised Semantic Segmentation Based on Co-segmentation28
This work proposes a novel method for weakly supervised semantic segmentation with only image-level labels, which relies on a large scale co-segmentation framework that can produce object masks for a group of images containing objects belonging to the same semantic class.
- 28
- 28
- 28
- 27
- 27
- 27
- 27
- 26
- 26
- PropMix: Hard Sample Filtering and Proportional MixUp for Learning with Noisy Labels25
This paper introduces the learning algorithm PropMix, which filters out hard noisy samples, and places clean and re-labelled easy noisy samples in a training set that is augmented with MixUp, removing the clean set size constraint and including a large proportion of correctly re-Labelledeasy noisy samples.
- MOLTR: Multiple Object Localization, Tracking and Reconstruction From Monocular RGB Videos25
MOLTR is presented, a solution to object-centric mapping using only monocular image sequences and camera poses that is able to localize, track and reconstruct multiple rigid objects in an online fashion when a RGB camera captures a video of the surrounding.
- 25
- 25
- 25
- Continual LLaVA: Continual Instruction Tuning in Large Vision-Language Models24
This work proposes a new benchmark for COntinuAl inStruction Tuning on LVLMs (COAST), which encompasses the aforementioned domain-incremental, capability-incremental, and dataset-incremental configurations and proposes Continual LLaVA, a rehearsal-free method tailored for continual instruction tuning in LVLMs.
- 24
- 24
- Structured Binary Neural Networks for Image Recognition23
This paper proposes a “group decomposition” strategy, termed GroupNet, which divides a network into desired groups, and extends the GroupNet for accurate semantic segmentation by embedding the rich context into the binary structure.
- Rotation Coordinate Descent for Fast Globally Optimal Rotation Averaging23
This paper presents a fast algorithm that achieves global optimality called rotation coordinate descent (RCD), which directly maintains and updates all valid rotations throughout the iterations and mathematically proves the convergence of the algorithm and empirically shows its superior efficiency over state-of-the-art global methods on a variety of problem configurations.
- Architecture Search of Dynamic Cells for Semantic Video Segmentation23
This work proposes a neural architecture search solution, where the choice of operations together with their sequential arrangement are being predicted by a separate neural network, and shows that such generalisation leads to stable and accurate results across common benchmarks, such as CityScapes and CamVid datasets.
- 23
- “Maximizing Rigidity” Revisited: A Convex Programming Approach for Generic 3D Shape Reconstruction from Multiple Perspective Views23
This work revisits the principle of “maximizing rigidity” in structure-from-motion literature, and develops a unified theory which is applicable to both rigid and non-rigid structure reconstruction in a rigidity-agnostic way.
- 23
- 23
- Novel View Specification and Synthesis23
A new method for directly specifying the novel camera motion for epipolar transfer is presented and a backward mapping scheme for trifocal transfer is developed which overcomes the problems associated with standard forward mapping methods.
- Learn to Predict Sets Using Feed-Forward Neural Networks22
This paper presents a novel approach for learning to predict sets with unknown permutation and cardinality using deep neural networks and demonstrates the validity of the set formulations on relevant vision problems such as multi-label image classification and a complex CAPTCHA test.
- Object Captioning and Retrieval with Natural Language22
This work proposes two new architectures to solve two related problems: object captioning and natural language-based object retrieval, and demonstrates that both problems can be solved effectively using hybrid end-to-end CNN-LSTM networks.
- Exploiting Temporal Information for DCNN-Based Fine-Grained Object Classification22
This work presents the novel task of video-based fine-grained object classification, proposes a corresponding new video dataset, and performs a systematic study of several recent deep convolutional neural network (DCNN) based approaches, which are specifically adapt to the task.
- 22
- Zooming while Tracking Using Affine Transfer22
The use of afne transfer is demonstrated to track while zooming, using clusters of corner features, a natural mechanism to allow features to appear and disappear while tracking, events which will occur as detail sharpens and dissolves during zooming.
- 21
- 21
- 21
- A Multiple View Layered Representation for Dynamic Novel View Synthesis21
A persistent representation of occupancy is maintained in spite of occlusion without enforcing a particular parametric shape model, and a MAP solution for estimating layer parameters which are consistent across views is formulated.
- 21
- Weakly Supervised Training of Monocular 3D Object Detectors Using Wide Baseline Multi-view Traffic Camera Data20
This work develops an alternative approach using a weakly supervised method of fine tuning 3D object detectors for traffic observation cameras; showing in the process that large existing autonomous vehicle datasets can be leveraged for pre-training.
- Probabilistic Parameter Selection for Learning Scene Structure from Video20
This work presents an online learning approach for robustly combining unreliable observations from a pedestrian detector to estimate the rough 3D scene geometry from video sequences of a static camera based on an entropy modelling framework.
- A Probabilistic Framework for Recognizing Similar Actions using Spatio-Temporal Features20
This paper proposes using an approximation of the full joint via first order dependencies between fe ature types using so-called Chow-Liu trees and achieves an improvement in the classification accuracy over naive Bayes an d other simple classifiers.
- Unconstrained Multiple-People Tracking20
This work presents two main contributions to achieve robust multiple-target tracking in uncontrolled scenarios: a novel system which consists on a hierarchical architecture which is devoted to one of the main tracking functionalities: target detection, low-level tracking, and high-level tasks such as target-appearance representation, or event management.
- 19
- Finding prototypes to estimate trajectory development in outdoor scenarios19
This work presents a method to analyze an existing set of trajectories obtained in a selected outdoor environment, clustered based on their entrance and exit points in order to model prototypes that are used to on–line estimate the development of a new trajectory.
- Joint Bayes Filter: A Hybrid Tracker for Non-rigid Hand Motion Recognition19
The novel contribution of the paper is that it unify the independent treatments of non-rigid motion and rigid motion into a robust Bayesian framework and can therefore be tracked without resorting to a complex kinematic model.
- 18
- 18
- Behaviour recognition and explanation for video surveillance18
This paper is concerned with producing high-level reports and explanations of human activity in video from a single, static camera in urban surveillance and sports video where the imaged person is medium/low resolution.
- 18
- 17
- 17
- Online unsupervised feature learning for visual tracking17
It is shown that a small dictionary, learned and updated online is as effective and more efficient than a huge dictionary learned offline, which facilitates the advantages of both feature learning and structured output prediction.
- 17
- 17
- 17
- Action Tokenizer Matters in In-Context Imitation Learning16
LipVQ-VAE, a variational autoencoder that enforces the Lipschitz condition in the latent action space via weight normalization, is proposed, which generates smoother actions and improves performance in high-fidelity simulators.
- 16
- Model Agnostic Saliency For Weakly Supervised Lesion Detection From Breast DCE-MRI16
The authors' model-agnostic 1-class saliency detector (MASD) is tested on weakly supervised breast lesion detection from DCE-MRI, achieving state-of-the-art detection accuracy when compared to current visualization methods.
- 16
- 16
- Joint Learning of Set Cardinality and State Distribution16
This work derives in a principled way, a mathematical formulation for set prediction where the output is permutation invariant, and jointly learns both the cardinality and the state distribution of the target set.
- Multiview segmentation and tracking of dynamic occluding layers16
An algorithm for the layered segmentation of video data in multiple views based on computing the parameters of a layered representation of the scene in which each layer is modelled by its motion, appearance and occupancy is presented.
- 16
- Automatic Human Behaviour Recognition and Explanation for CCTV Video Surveillance16
This paper is concerned with producing high-level text reports and explanations of human activity in video from a single, static camera to enable surveillance analysts to maintain situational awareness despite the presence of large volumes of data.
- 16
- Self-alignment of a binocular robot16
A novel method by which a four-axis binocular head platform can autonomously align its cameras so that their optic axes are parallel to each other and to the forward direction of the robot is described.
- 15
- 15
- 14
- 14
- Cooperative surveillance of multiple targets using mutual information14
A method to control multiple, but diverse pan-tilt-zoom cameras which are sharing overlapping views of the same spatial location for the purpose of observation of this scene is presented, obviating the need for a wide-angle supervisor camera or other artificial restrictions on the camera parameter settings.
- An Evaluation of Shape Descriptors for Image Retrieval in Human Pose Estimation14
It is suggested that a simple linear transformation of the image (such as the DCT) is as effective as the more complex, non-linear methods in the task of human pose estimation.
- 14
- CropMix: Sampling a Rich Input Distribution via Multi-Scale Cropping13
This work demonstrates that CropMix can be seamlessly applied to virtually any training recipe and neural network architecture performing classification tasks, and is of benefit to both contrastive learning and masked image modeling towards more powerful representations.
- 13
- Low-Rank Kernel Subspace Clustering.13
This paper proposes to learn a low-rank kernel mapping, with which the mapped data in feature space are not only low- rank but also self-expressive, such that the low-dimensional subspace structures are present and manifested in the high-dimensional feature space.
- Improving Global Multi-target Tracking with Local Updates13
This work proposes a scheme to explicitly detect and resolve ambiguous situations in multiple target tracking, and proposes a local data association scheme to maintain the target identities in these situations when targets become occluded.
- Trajectory alignment and evaluation in SLAM: Horn’s method vs alignment on the manifold13
The Problem of Mobile Sensors: Setting future goals and indicators of progress for SLAM Workshop in conjunction with RSS 2015 is discussed.
- Camera calibration from human motion13
A method for the self-calibration of non-rigid affine structure to a Euclidean co-ordinate frame from only two views by enforcing constraints derived from the known structure of the human body, such as piecewise rigidity and approximate symmetry is presented.
- Automated Alignment of Robotic Pan-Tilt Camera Units Using Vision12
This paper shows how to carry out an automatic alignment of a pan-tilt camera platform with its natural coordinate frame, using only images obtained from the cameras during controlled motion of the unit, with the advantages of applicability to a wide range circumstances and no need to involve calibration objects or complex motions.
- 12
- TANGO: Traversability-Aware Navigation with Local Metric Control for Topological Goals11
This work presents a novel RGB-only, object-level topometric navigation pipeline that enables zero-shot, long-horizon robot navigation without requiring 3D maps or pre-trained controllers, and outperforms existing state-of-the-art methods.
- 11
- 11
- 11
- 11
- A discrete-time attitude observer on SO(3) for vision and GPS fusion11
Through experimental studies, it is demonstrated that the observer effectively compensates for the inherent drift of the pure monocular vision based attitude estimation and is able to recover the North-East-Down orientation even if it is initialized with a very large attitude error.
- 11
- Steering without Representation with the Use of Active Fixation11
It is suggested that proportional rule devised for steering in the robotic example appears applicable to the observed human performance data, providing an alternative explanation to the quadratic rule proposed by Land and Lee.
- Meta Learning with Differentiable Closed-form Solver for Fast Video Object Segmentation10
This paper tackles the task of fast segmenting all pixels of a target object in all frames, given the annotation mask in the first frame, as a meta-learning problem, where the base learner grasping the semantic scene understanding for a general type of objects, and the meta learner quickly adapting the appearance of the target object with a few examples.
- How Trustworthy are the Existing Performance Evaluations for Basic Vision Tasks10
This paper examines performance evaluation criteria for basic vision tasks namely, object detection, instance-level segmentation and multi-object tracking, and advocates the use of criteria that are consistent with mathematical requirements, contextually meaningful in sanity tests, and robust to hyper-parameters for reliability.
- Template-Based Automatic Search of Compact Semantic Segmentation Architectures10
This work proposes a novel solution able to find light-weight and accurate segmentation architectures starting from only few blocks of a pre-trained classification network that relies on templates of sets of operations, predicts which template and how many times should be applied at each step, while also generating the connectivity structure and downsampling factors.
- 10
- 10
- 10
- Learning Multi-level Region Consistency with Dense Multi-label Networks for Semantic Segmentation10
This work proposes a dense multi-label network module that is able to encourage the region consistency at different levels and can be easily integrated into any semantic segmentation systems.
- Computer Vision -- ACCV 201410
An automatic vessel segmentation framework which can achieve highly accurate segmentation even in regions of low contrast and signal-to-noise-ratios (SNRs) and at vessel boundaries with disturbance induced by adjacent non-vessel pixels is described.
- Model-based recognition and range imaging for a guided vehicle10
Progress in providing the Oxford Autonomous Guided Vehicle (AGV) with range sensing and model-based recognition of classes of industrial artifacts is described and a system for pose recognition of generic models from range-imagery is described.
- 9
- Binary Constrained Deep Hashing Network for Image Retrieval Without Manual Annotation9
A novel end-to-end deep learning approach for the image retrieval task, in which the network is trained to produce binary codes directly from image pixels without the need for manual annotation, and a novel pairwise constrained loss function which simultaneously encodes the distances between pairs of hash codes.
- 9
- 9
- 9
- Understanding Interactions and Guiding Visual Surveillance by Tracking Attention9
A fully automatic method to determine a person's attention based on real-time visual tracking of their heads and a coarse classification of their head pose is described.
- What Images are More Memorable to Machines?8
The concept of machine memorability is proposed and opens a new research direction at the interface between machine memory and visual data.
- Visual localization under appearance change: filtering approaches8
It is shown that exploiting temporal continuity in the testing sequence significantly improves visual localization and achieves better results than state of the art approaches for the task on visual localization under significant appearance change.
- Towards Semantic SLAM: Points, Planes and Objects8
Experiments show the proposed points-planes-quadrics representation can easily incorporate Manhattan and object affordance constraints, greatly improving camera localization and leading to semantically meaningful maps.
- 8
- 8
- Image interpolation for virtual sports scenarios8
The EVENTS project attempts to apply state of the art view interpolation to the field of professional sports to populate a wide scenario such as a stadium with a number of cameras and, via computer vision, to produce photo-realistic moving or static images from virtual viewpoints.
- Binocular Self-Alignment and Calibration from Planar Scenes8
This work considers the problem of aligning and calibrating a binocular pan-tilt device using visual information from controlled motions, while viewing a degenerate (planar) scene, and shows how to construct the images of points at infinity in various visual directions.
- GraspMamba: A Mamba-based Language-driven Grasp Detection Framework with Hierarchical Feature Learning7
GraspMamba is introduced, a new language-driven grasp detection method that employs hierarchical feature fusion with Mamba vision to tackle challenges of cluttered images, lengthy textual descriptions, or slow inference speed.
- Autonomy and Perception for Space Mining7
The solution for Phase 2 of the NASA Space Robotics Challenge provided a simulated lunar environment in which teams were tasked to develop software systems to achieve autonomous collaborative robots for mining on the Moon, showing how machine learning-enabled vision could alleviate major challenges posed by the lunar environment towards autonomous space mining.
- 7
- Learning Deeply Supervised Good Features to Match for Dense Monocular Reconstruction7
This work proposes a novel convolutional neural network architecture along with a deeply supervised feature learning scheme for pixel-wise regression of visual descriptors from an image which are best suited for dense monocular SLAM.
- 7
- The Rôle of Self-Calibration in Euclidean Reconstruction from Two Rotating and Zooming Cameras7
An analysis of near-ambiguities which commonly arise in the self-calibration of rotating cameras and how their effects may be mitigated by exploiting the epipolar geometry is provided.
- Recognition of parameterized objects from 3D data: a parallel implementation7
A parallel implementation of an algorithm designed to recognize fully parameterized objects from 3D data through the exploitation of the intrinsic parallelism of the SUP/INF network, while the interpretation tree search is also distributed on the transputer based MIMD architecture.
- 6
- HM$^4$: Hidden Markov Model With Memory Management for Visual Place Recognition6
This work develops a Hidden Markov Model approach with a two-tiered memory management that allows constant time and space inference for a fixed coverage area of visual place recognition in autonomous driving and derives a compact image representation inspired by Locality Sensitive Hashing.
- Visual Localization under Appearance Change: A Filtering Approach6
This paper shows that exploiting temporal continuity in the testing sequence significantly improves visual localization — qualitatively and quantitatively — and proposes an image retrieval pipeline which relies on local features and an encoding technique to represent an image as a single vector.
- 6
- A branch-and-bound algorithm for checkerboard extraction in camera-laser calibration6
This work addresses the problem of camera-to-laserscanner calibration using a checkerboard and multiple imagelaser scan pairs and proposes a branch-and-bound technique that deterministically and globally optimizes the objective.
- 6
- 6
- 6
- Transfer of Fixation Using Affine Structure: Extending the Analysis to Stereo6
It is shown that fixation and explicit structure recovery can occur separately, allowing the information required for gaze control to be computed in a fixed time.
- Active exploration of dynamic and static scenes6
The construction of an agile camera platform, a highbandwidth parallel vision system, and several real-time motion and vergence processes that enable an active camera to perform motion saccades to “capture” moving targets, and thereafter to track them are described.
- Social-MAE: Social Masked Autoencoder for Multi-Person Motion Representation Learning5
Social-MAE, a simple yet effective transformer-based masked autoencoder framework for multi-person human motion data, uses masked modeling to pre-train the encoder to reconstruct masked human joint trajectories, enabling it to learn generalizable representations of motion in human crowded scenes.
- 5
- Sensor Allocation and Online-Learning-Based Path Planning for Maritime Situational Awareness Enhancement: A Multi-Agent Approach5
Two algorithms are presented: a greedy algorithm for assigning sensors to targets, and a distributed multi-agent path planning algorithm based on regret-matching learning based on regret-matching learning that achieves situational awareness more quickly than related work.
- 5
- 5
- Optimizable Object Reconstruction from a Single View.5
This framework uses the power of deep learning to capture 3D shape information from training data and provide high-quality initialization, while allowing both image evidence and shape priors to influence iterative refinement at inference time.
- Regressing Local to Global Shape Properties for Online Segmentation and Tracking5
A novel regression based framework that uses online learned shape information to reconstruct occluded object contours by representing shapes using their 2D discrete cosine transforms and by regressing low frequency from high frequency harmonics.
- 5
- 5
- Action recognition using shared motion parts5
It is shown that the inclusion of motion information significantly improves the recognition accuracy, and is tested using a standard action database containing 10 action classes obtaining perfect classification.
- 5
- Real-time Visual Recovery of Pose using Line Tracking in Multiple Cameras5
The system, which uses three cameras a low-end commercial framegrabber and runs on single PC, has been devised to provide simple visual feedback for the teleoperator of a force reecting robot manipulator.
- G3Splat: Geometrically Consistent Generalizable Gaussian Splatting4
G3Splat is introduced, a geometry-consistent generalizable splatting framework that addresses degeneracies through differentiable geometric priors on the predicted 3D Gaussians, making the learning problem well-posed.
- Seeing through Imagination: Learning Scene Geometry via Implicit Spatial World Modeling4
MILO is introduced, an Implicit spatio-spatial reasoning modeling paradigm that simulates human-like spatial imagination and proposes RePE (Relative Positional Encoding), a novel encoding scheme that captures relative camera-pose transformations, offering superior performance over absolute coordinate systems.
- 4
- 4
- Learning Deeply Supervised Visual Descriptors for Dense Monocular Reconstruction.4
This work proposes a novel convolutional neural network architecture along with a deeply supervised feature learning scheme for pixel-wise regression of visual descriptors from an image which are best suited for dense monocular SLAM.
- 4
- From Visual Query to Visual Portrayal4
This paper introduces a method to fuse the rendered novel views from all input images at each virtual view point by computing their intrinsic image and illuminations and demonstrates the performance of the prototype system on two landmark structures.
- Language and Planning in Robotic Navigation: A Multilingual Evaluation of State-of-the-Art Models3
This study presents the first-ever work in Arabic language integration within the Vision-and-Language Navigation (VLN) domain in robotics, an area that has been notably underexplored in existing research.
- 3
- Improved Visual Localization via Graph Filtering3
This work introduces a framework which aims at enhancing the performance of retrieval-based localization methods by taking into account additional information available, such as GPS coordinates or temporal proximity in the acquisition of the images.
- Deep Reinforcement Learning for Detecting Breast Lesions from DCE-MRI3
A detection model that is capable of accelerating the inference time of lesion detection from breast dynamically contrast-enhanced magnetic resonance images (DCE-MRI) at state-of-the-art accuracy is presented and reaches a true positive rate (TPR) of 0.8 at around three false positive detections.
- 3
- Temporal Priors for Novel Video Synthesis3
A method to construct a virtual sequence for a camera moving through a static environment given an input sequence from a different camera trajectory, and defines an MRF on the graph which encodes photoconsistency of pixels as well as texture priors in both space and time.
- 3
- The Pipe-group Architecture for Real-Time Active Vision3
The paper illustrates the use of the pipe-group processing architecture and its real-time processing capabilities, with an example of active stereo tracking and structure recovery using the method of affine tranfer.
- Constructing front-ends to existing software systems3
The FOCUS project is concerned with the development of tools and methods to aid in the construction of knowledge-based front-ends to existing software systems to extend the life and usability of valuable code that could be more widely used.
- Recognition of Parametrised Models from 3D Data.3
This paper addresses the problem of recognising an object from range-data observations as an instance of a parametric model class, and determining the values of the class parameters for that instance of the model, and the pose of the object.
- Numerical problem-solving environments — current and future trends3
This paper looks at the work being carried out at NAG Limited, in Oxford, towards producing Numerical Problem-Solving Environments, discussing systems that are currently available and those for which to be striving.
- BEVPose: Unveiling Scene Semantics through Pose-Guided Multi-Modal BEV Alignment2
This work presents BEVPose, a framework that integrates BEV representations from camera and lidar data, using sensor pose as a guiding supervisory signal, and demonstrates promising performance in BEV map segmentation tasks, while necessitating only a minimal amount of annotated data.
- Weakly Supervised Test-Time Domain Adaptation for Object Detection2
The proposed method only requires the operator to provide weak labels, which are then used to guide the adaptation process, and it is shown that the proposed method outperforms existing works, demonstrating a great benefit of human-in-the-loop test-time domain adaptation.
- Symmetry-Breaking Augmentations for Ad Hoc Teamwork2
SBA encourages agents to learn robust responses to unknown strategies, highlighting how social conventions impact human-AI alignment and provides insights into how AI systems can better adapt to diverse human conventions and the core mechanics of alignment.
- Semantic Segmentation on 3D Point Clouds with High Density Variations2
A novel architecture called HDVNet is proposed that contains a nested set of encoder-decoder pathways, each handling a specific point density range, and outperforms state-of-the-art models in segmentation accuracy on real point clouds with inconsistent density.
- The Edge of Disaster: A Battle Between Autonomous Racing and Safety2
This work provides a model predictive control baseline that is able to, in a single lap, safely adapt to an unseen racetrack and achieve competitive lap times.
- 2
- 2
- 2
- 2
- 2
- Self-Calibration of Zooming Cameras from a Single Viewpoint2
A configuration that commonly occurs in many imaging applications is that in which a camera has its optic centre fixed in location but which may rotate and zoom, thus changing its internal parameters.
- 2
- 1
- 1
- 1
- 1
- 1
- 1
- Practical Visual Localization for Autonomous Driving: Why Not Filter?1
A novel particle filtering technique that works in conjunction with a visual localization method to achieve accurate city-scale localization that is robust against environmental variations that is convincing on synthetic and real datasets.
- Rethinking Binary Neural Network for Accurate Image Classification and Semantic Segmentation1
This paper proposes to train a network with both binary weights and binary activations, designed specifically for mobile devices with limited computation capacity and power consumption, and claims that considering both value and structure approximation should be the future development direction of BNNs.
- Lesion Saliency for Weakly Supervised Lesion Detection from Breast DCE-MRI1
The model-agnostic 1-class saliency detector (MASD) is tested on weakly supervised breast lesion detection from DCE-MRI, achieving state-of-the-art detection accuracy when compared to current visualization methods.
- Non-Linear Subspace Clustering with Learned Low-Rank Kernels1
This paper proposes to learn a low-rank kernel matrix, with which mapped data in feature space are not only low- rank but also self-expressive, and the low-dimensional subspace structures of the (implicitly) mapped data are retained and manifested in the high-dimensional feature space.
- Image Matching Benchmark1
This work presents the first application-oriented image matching benchmark to facilitate the analysis of matching algorithms in application level and demonstrates the effectiveness of a simple technique which is readily pluggable into any matching system to improve performance.
- Mean-Shift Visual Tracking with NP-Windows Density Estimates1
This paper applies a general non-parametric PDF estimation method, namely NP-Windows, to replace the histogram estimating the colour probabilities in mean-shift visual tracking, and shows that this leads to faster and more accurate convergence, and by making use of a (separable) Gaussian kernel rather than the more usual Epanechnikov kernel, that the actual run-time can be reduced.
- 1
- 1
- Stereo Fixation using Affne Transfer1
An algorithm which uses affine transfer of the fixation point in a stereo pair to obtain stereo fixation of a moving object is described, providing immunity to poor choice of basis features.
- 1
- What do we mean by expert systems?1
The purpose of this paper is to expose readers to the concept and vocabulary of expert systems, especially in the context of scientific software, and to introduce a number of related topics such as knowledge-based systems.
- –
- Simultaneous Diffusion Sampling for Conditional LiDAR generation–
This paper proposes a novel simultaneous diffusion sampling methodology to generate point clouds conditioned on the 3D structure of the scene as seen from multiple views, allowing it to outperform existing methods by a large margin in a variety of benchmarks.
- Category Level 6D Object Pose Estimation from a Single RGB Image using Diffusion–
A unified generative framework that addresses both single-view category-level pose estimation and temporal sequence tracking using only RGB input is introduced and it is demonstrated that the iterative nature of the score-based sampler enables a seamless transition to video sequences by preserving and propagating the multi-hypothesis distribution across time as a coherent temporal prior.
- Few and Fewer: Learning Better from Few Examples Using Fewer Base Classes–
This paper investigates whether better features for the target dataset can be obtained by training on fewer base classes, seeking to identify a more useful base dataset for a given task.
- Oops, I Sampled it Again: Reinterpreting Confidence Intervals in Few-Shot Learning–
This research demonstrates that the use of paired tests can partially address the issue of misleading confidence intervals in FSL comparative studies and explores methods to further reduce the (size of the) CI by strategically sampling tasks of a specific size.
- –
- –
- Predicting Topological Maps for Visual Navigation in Unexplored Environments–
This work describes a navigation system that uses the layout predictions to satisfy high-level goals more rapidly and accurately than the prior art and test the framework in Matterport3D to show more success and efficient navigation in unseen environments.
- –
- –
- –
- Auxiliary tasks to improve trip hazard affordance detection on construction sites–
This work proposes to train a CNN performing pixel-wise trip detection with three auxiliary tasks to help the CNN better infer scene geometric properties of trip hazards, and makes it plausible to deploy a robotic platform to perform trip hazard detection, and so potentially reduce the number of injuries on construction sites.
- –
- –
- Temporally-coherent Novel Video Synthesis Using Texture-based Priors–
This paper proposes a new spatio-temporal approach to novel video synthesis that exploits epipolar geometry to impose constraints on temporal coherence of the rendered views, and defines an MRF on the graph which encodes photoconsistency of pixels as well as texture priors in both space and time.
- –
- Numerical algorithm for metrology.–
This talk will look briefly at the METROS Environment and most importantly how it can be best populated with solution software.
- –
- Recognizing Parameterized Models Using 3D Edges–
An object’s pose and class parameters are recovered by the algorithm using 3D edge fragments extracted from a stereo pair of images.
- Interfaces to numerical packages–
Some of the work with which NAG Limited is involved, in the development of knowledge-based front-ends for existing numerical and statistical software, is described.
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.