Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks292 from public data
- Ego4D: Around the World in 3,000 Hours of Egocentric Video2,124
Ego4D, a massive-scale egocentric video dataset and benchmark suite, is introduced and a host of new benchmark challenges centered around understanding the first-person visual experience in the past, present, and future are presented.
- Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking1,056
This work shows that a basic Kalman filter can still obtain state-of-the-art tracking performance if proper care is taken to fix the noise accumulated during Occlusion, and improves robustness during occlusion and non-linear motion.
- AgentFormer: Agent-Aware Transformers for Socio-Temporal Multi-Agent Forecasting792
A stochastic multi-agent trajectory prediction model that can attend to features of any agent at any previous timestep when inferring an agent’s future position is proposed and significantly improves the state of the art on well-established pedestrian and autonomous driving datasets.
- Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectives640
To push the frontier of first-person video understanding of skilled human activity, a suite of benchmark tasks and their annotations are presented, including fine-grained activity understanding, proficiency estimation, cross-view translation, and 3D hand/body pose.
- Activity Forecasting639
The unified model uses state-of-the-art semantic scene understanding combined with ideas from optimal control theory to achieve accurate activity forecasting and shows how the same techniques can improve the results of tracking algorithms by leveraging information about likely goals and trajectories.
- 597
- No-Reference Image Quality Assessment via Transformers, Relative Ranking, and Self-Consistency481
A novel model that leverages self-consistency as a source of self-supervision to improve the robustness of NR-IQA models is proposed and enforce self- Consistency between the outputs of the quality assessment model for each image and its transformation to reduce the uncertainty of the model.
- DanceTrack: Multi-Object Tracking in Uniform Appearance and Diverse Motion467
A large-scale dataset for multi-human tracking, where humans have sim-ilar appearance, diverse motion and extreme articulation, and is named “DanceTrack”, which is expected to provide a better platform to develop more MOT algorithms that rely less on visual discrimination and depend more on motion analysis.
- 434
- PRECOG: PREdiction Conditioned on Goals in Visual Multi-Agent Settings419
A probabilistic forecasting model of future interactions between a variable number of agents that performs both standard forecasting and the novel task of conditional forecasting, which reasons about how all agents will likely respond to the goal of a controlled agent.
- OmniH2O: Universal and Dexterous Human-to-Humanoid Whole-Body Teleoperation and Learning417
An RL-based sim-to-real pipeline, which involves large-scale retargeting and augmentation of human motion datasets, learning a real-world deployable policy with sparse sensor input by imitating a privileged teacher policy, and reward designs to enhance robustness and stability is developed.
- Rethinking Transformer-based Set Prediction for Object Detection389
Experimental results show that the proposed methods not only converge much faster than the original DETR, but also significantly outperform DETR and other baselines in terms of detection accuracy.
- Deep OC-Sort: Multi-Pedestrian Tracking by Adaptive Re-Identification381
A novel way to leverage objects’ appearances to adaptively integrate appearance matching into existing high-performance motion-based methods, building upon the pure motion- based method OC-SORT.
- Perpetual Humanoid Control for Real-time Simulated Avatars365
A physics-based humanoid controller is presented that achieves high-fidelity motion imitation and fault-tolerant behavior in the presence of noisy input and unexpected falls and dynamically allocates new network capacity to learn harder and harder motion sequences.
- Cross-Domain Adaptive Teacher for Object Detection329
A teacher-student framework named Adaptive Teacher (AT) which leverages domain adversarial learning and weak-strong data augmentation to address the domain gap in object detection and demonstrates superiority over existing approaches and even Oracle (fully-supervised) models by a large margin.
- Going Deeper into First-Person Activity Recognition318
By learning to recognize objects, actions and activities jointly, the performance of individual recognition tasks also increase by 30% (actions) and 14% ( objects) and the results of extensive ablative analysis are included to highlight the importance of network design decisions.
- DLow: Diversifying Latent Flows for Diverse Human Motion Prediction315
A novel sampling method, Diversifying Latent Flows (DLow), to produce a diverse set of samples from a pretrained deep generative model that outperforms state-of-the-art baseline methods in terms of sample diversity and accuracy.
- Learning Human-to-Humanoid Real-Time Whole-Body Teleoperation311
This work presents Human to Humanoid (H2O), a reinforcement learning (RL) based framework that enables real-time whole-body teleoperation of a full-sized humanoid robot with only an RGB camera, and achieves teleoperation of dynamic whole-body motions in real-world scenarios.
- 307
- Monocular 3D Object Detection with Pseudo-LiDAR Point Cloud296
This work aims at bridging the performance gap between 3D sensing and 2D sensing for 3D object detection by enhancing LiDAR-based algorithms to work with single image input by enhancing pseudo-LiDAR end-to-end methods.
- NavCog288
This work proposes a smartphone-based system that provides turn-by-turn navigation assistance based on accurate real-time localization over large spaces and shows that the system is capable of guiding visually impaired users in complex and unfamiliar environments.
- 277
- 265
- Time Will Tell: New Outlooks and A Baseline for Temporal Multi-View 3D Object Detection260
The theoretical and empirical analysis demonstrates that the optimal temporal difference between views varies significantly for different pixels and depths, making it necessary to fuse many timesteps over long-term history to generate a cost volume from a long history of image observations, compensating for the coarse but efficient matching resolution with a more optimal multi-view matching setup.
- GNN3DMOT: Graph Neural Network for 3D Multi-Object Tracking With 2D-3D Multi-Feature Learning259
This work proposes two techniques to improve the discriminative feature learning for MOT by introducing a novel feature interaction mechanism by introducing the Graph Neural Network and proposes a novel joint feature extractor to learn appearance and motion features from 2D and 3D space simultaneously.
- r2p2: A ReparameteRized Pushforward Policy for Diverse, Precise Generative Path Forecasting259
A method to forecast a vehicle’s ego-motion as a distribution over spatiotemporal paths, conditioned on features embedded in an overhead map, and obtains expressions for the cross-entropy metrics that can be efficiently evaluated and differentiated, enabling stochastic-gradient optimization.
- Deep Supervised Hashing with Triplet Labels241
A triplet label based deep hashing method which aims to maximize the likelihood of the given triplet labels, which outperforms all the baselines on CIFAR-10 and NUS-WIDE datasets, including the state-of-the-art method DPSH.
- CaBot: Designing and Evaluating an Autonomous Navigation Robot for Blind People198
The design of CaBot (Carry-on roBot), an autonomous suitcase-shaped navigation robot that is able to guide blind users to a destination while avoiding obstacles on their path is presented.
- Trace and Pace: Controllable Pedestrian Animation via Guided Trajectory Diffusion196
A method for generating realistic pedestrian trajectories and full-body animations that can be controlled to meet user-defined goals and utilizing the value function learned during RL training of the animation controller to guide diffusion to produce trajectories better suited for particular scenarios such as collision avoidance and traversing uneven terrain is introduced.
- GLAMR: Global Occlusion-Aware Human Mesh Recovery with Dynamic Cameras194
The proposed approach outperforms prior methods significantly in terms of motion infilling and global mesh recovery and presents a global optimization framework that refines the predicted trajectories and optimizes the camera poses to match the video evidence such as 2D keypoints.
- Universal Humanoid Motion Representations for Physics-Based Control192
A universal motion representation that encompasses a comprehensive range of motor skills for physics-based humanoid control and demonstrates the effectiveness of the motion representation by solving generative tasks and motion tracking using VR controllers.
- SimPoE: Simulated Character Control for 3D Human Pose Estimation173
SimPoE, a Simulation-based approach for 3D human Pose Estimation, which integrates image-based kinematic inference and physics-based dynamics modeling, is presented and a meta-control mechanism that dynamically adjusts the character’s dynamics parameters based on the character state to attain more accurate pose estimates is proposed.
- 3D Human Motion Estimation via Motion Compression and Refinement172
A technique for generating smooth and accurate 3D human pose and motion estimates from RGB video sequences using auto-encoder-based motion compression and a residual representation learned through motion refinement.
- NavCog3172
NavCog3, a smartphone-based indoor navigation assistant that has been evaluated in a 21,000 m2 shop- ping mall, is implemented and suggested to open a new opportunity for users with visual impairments to independently find and visit large and complex places with confidence.
- 167
- Ego-Pose Estimation and Forecasting As Real-Time PD Control152
This work proposes the use of a proportional-derivative control based policy learned via reinforcement learning (RL) to estimate and forecast 3D human pose from egocentric videos and introduces a value function based fail-safe mechanism which enables the method to run as a single pass algorithm over the video data.
- First-Person Activity Forecasting with Online Inverse Reinforcement Learning148
DARKO learns and forecasts from first-person visual observations of the user’s daily behaviors via an Online Inverse Reinforcement Learning (IRL) approach, and shows DARKO forecasts goals better than competing methods in both noisy and ideal settings.
- 147
- People with Visual Impairment Training Personal Object Recognizers141
This work adopts transfer learning with a deep learning system for user-defined multi-label k-instance classification and explores personal object recognizers, where visually impaired people train a mobile application with a few snapshots of objects of interest and provide custom labels.
- IDOL: Inertial Deep Orientation-Estimation and Localization133
A two-stage, data-driven pipeline using a commodity smartphone that first estimates device orientations and then estimates device position outperforms state-of-the-art methods in both orientation and position error on a large dataset that contains 20 hours of pedestrian motion across 3 buildings and 15 subjects.
- A Baseline for 3D Multi-Object Tracking132
This work proposes a simple yet accurate real-time baseline 3D MOT system, using an off-the-shelf 3D object detector to obtain oriented 3D bounding boxes from the LiDAR point cloud and using a combination of 3D Kalman filter and Hungarian algorithm for state estimation and data association.
- Twitter A11y: A Browser Extension to Make Twitter Images Accessible130
Twitter A11y increases access to social media platforms for people with visual impairments by providing high-quality automatic descriptions for user-posted images by increasing alt-text coverage from 7.6% to 78.5%, before crowdsourcing descriptions for the remaining images.
- Action-Reaction: Forecasting the Dynamics of Human Interaction130
This work model dual-agent interactions as an optimal control problem, where the actions of the initiating agent induce a cost topology over the space of reactive poses – a space in which the reactive agent plans an optimal pose trajectory.
- BBeep129
An assistive suitcase system for supporting blind people when walking through crowded environments using pre-emptive sound notifications, BBeep, and it is observed that the proposed system significantly reduces the number of imminent collisions.
- 122
- RePOSE: Fast 6D Object Pose Refinement via Deep Texture Rendering110
RePOSE leverages image rendering for fast feature extraction using a 3D model with a learnable texture and utilizes differentiable Levenberg-Marquardt (LM) optimization to refine a pose fast and accurately by minimizing the distance between the input and rendered image representations without the need of zooming in.
- Virtual Navigation for Blind People109
Two smartphone-based virtual navigation interfaces are presented: VirtualLeap, which allows the user to jump through a sequence of street intersection labels, turn-by-turn instructions and POIs along the route; and VirtualWalk, which simulates variable speed step-by -step walking using audio effects, whilst conveying similar route information.
- Airport Accessibility and Navigation Assistance for People with Visual Impairments102
This study presents the first systematic evaluation posing BLE technology as a strong approach to increase the independence of visually impaired people in airports and finds that despite the challenging environment participants were able to complete their itinerary independently.
- “It's almost like they're trying to hide it”: How User-Provided Image Descriptions Have Failed to Make Twitter Accessible102
It is found that simply making it possible to provide image descriptions is not enough, and future directions for automated tools that may support users in writing high-quality descriptions are revealed.
- 101
- 95
- 95
- NavCog3 in the Wild94
To assess the capability of NavCog3 to promote independent mobility of individuals with visual impairments, the system was deployed and evaluated in two challenging real-world scenarios and the system’s usability in the wild was validated in a hotel complex temporarily equipped with NavCogs3 during a conference for individuals withvisual impairments.
- Understanding Hand-Object Manipulation with Grasp Types and Object Attributes94
The hypothesis is that it is necessary to model the grasp types of hands and the attributes of manipulated objects in order to accurately recognize manipulation actions, and the proposed action model outperforms traditional appearance-based models which are not designed to take into account semantic constraints such as grasp types or object attributes.
- Occluded Human Mesh Recovery93
Occluded Human Mesh Recovery (OCHMR) is presented - a novel top-down mesh recovery approach that incorporates image spatial context to overcome the limitations of the single-human assumption and achieves superior performance on challenging multi-person benchmarks like 3DPW, CrowdPose and OCHuman.
- Back-Hand-Pose90
This work introduces DorsalNet, a two-stream convolutional neural network to regress finger joint angles from spatio-temporal features of the dorsal hand region (the movement of bones, muscle, and tendons), which is the first vision-based real-time 3D hand pose estimator using visual features from the dorsalhand region.
- Softer-NMS: Rethinking Bounding Box Regression for Accurate Object Detection.90
A novel boundingbox regression loss for learning bounding box transformation and localization variance together is introduced, which exhibits a strong connection to localization accuracy, which is then utilized in this new non-maximum suppression method to improve localization accuracy for object detection.
- Optical Non-Line-of-Sight Physics-Based 3D Human Pose Estimation87
This work brings together a diverse set of technologies from NLOS imaging, human pose estimation and deep reinforcement learning to construct an end-to-end data processing pipeline that converts a raw stream of photon measurements into a full 3D human pose sequence estimate.
- 85
- 85
- 84
- 3D Ego-Pose Estimation via Imitation Learning81
This work proposes a novel control-based approach to model human motion with physics simulation and uses imitation learning to learn a video-conditioned control policy for ego-pose estimation and allows for domain adaption to transfer the policy trained on simulation data to real-world data.
- EgoHumans: An Egocentric 3D Multi-Human Benchmark78
EgoFormer, a novel approach with a multi-stream transformer architecture and explicit 3D spatial reasoning to estimate and track the human pose, is proposed, which significantly outperforms prior art on the EgoHumans dataset.
- 77
- EdgeSonic73
Preliminary experiments show that the combination of local edge gradient sonification and distance-to-edge sonification are effective for understanding basic line drawings and a significant improvement in image understanding with the introduction of proper user training.
- BFM-Zero: A Promptable Behavioral Foundation Model for Humanoid Control Using Unsupervised Reinforcement Learning71
A first-of-its-kind model that learns an effective shared latent representation that embeds motions, goals, and rewards into a common space, enabling a single policy to be prompted for multiple downstream tasks without retraining, BFM-Zero establishes a step toward scalable, promptable behavioral foundation models for whole-body humanoid control.
- Virtual navigation for blind people: Transferring route knowledge to the real-World71
An interactive virtual navigation app where users can learn unfamiliar routes before physically visiting the environment and analysis of the evolution of route knowledge found that participants were able to quickly learn shorter routes and gradually increase their knowledge in both short and long routes.
- Privacy-Preserving Visual Learning Using Doubly Permuted Homomorphic Encryption71
The proposed privacy-preserving framework is designed to aggregate multiple classifiers updated locally using private data and to ensure that no private information about the data is exposed during and after its learning procedure, using a homomorphic cryptosystem that can aggregate the local classifiers while they are encrypted and thus kept secret.
- Smartphone-based localization for blind navigation in building-scale indoor environments70
A series of techniques that enhance a probabilistic localization algorithm that utilizes mobile device inertial sensors and Received Signal Strength (RSS) from Bluetooth Low Energy (BLE) beacons are presented.
- Achieving Practical and Accurate Indoor Navigation for People with Visual Impairments69
This work improves over a traditional regression- based localization approach by introducing a novel, graph-based localization method using Pedestrian Dead Reckoning (PDR) and particle filter, and studies with users show the impact that the increased accuracy has on the usability of the navigation application for the visually impaired.
- 69
- Omnigrasp: Grasping Diverse Objects with Simulated Humanoids68
A method for controlling a simulated humanoid to grasp an object and move it to follow an object's trajectory to leverage a humanoid motion representation that provides human-like motor skills and significantly speeds up training.
- ReCog: Supporting Blind People in Recognizing Personal Objects68
It is shown that ReCog enables blind users to train and recognize their personal objects, and that camera-aiming guidance helps novice users to increase their confidence, achieve better accuracy, and learn strategies to capture better photos.
- 67
- 63
- The Present and Future of Museum Accessibility for People with Visual Impairments63
This paper surveyed 19 PVI to understand their opinions and expectations about visiting museums independently, as well as the requirements of user interfaces to support it, and increases the knowledge about the previous experiences, motivations and accessibility issues of PVI in museums.
- HARMONIC: A multimodal dataset of assistive human–robot collaboration62
A large multimodal dataset of human interactions with a robotic arm in a shared autonomy setting designed to imitate assistive eating, which could be of interest to researchers studying intention prediction, human mental state modeling, and shared autonomy.
- 62
- StereOBJ-1M: Large-scale Stereo Image Dataset for 6D Object Pose Estimation61
A novel method for efficiently annotating pose data in a multi-view fashion that allows data capturing in complex and flexible environments and a novel object-level pose optimization method for computing 6D pose from keypoint predictions in multiple images are proposed.
- 61
- G-HOP: Generative Hand-Object Prior for Interaction Reconstruction and Grasp Synthesis60
G-HOP, a denoising diffusion based generative prior for hand-object interactions that allows modeling both the 3D object and a human hand, conditioned on the object category, represents a first approach that allows jointly generating both hand and object.
- 59
- NavCog57
The accurate localization achieved by NavCog is used for precise turn-by-turn way-finding assistance as the first step, but the ultimate goal is to present a variety of location based information to the user, such as points of interest gathered from social media and online geografic information services.
- 56
- AttentionNAS: Spatiotemporal Attention Cell Search for Video Classification54
This work proposes a novel search space for spatiotemporal attention cells, which allows the search algorithm to flexibly explore various design choices in the cell, and demonstrates strong generalization across different modalities, backbones, and datasets.
- Domain Randomization for Scene-Specific Car Detection and Pose Estimation53
This work proposes to solve domain gap through the use of appearance randomization to generate a wide range of synthetic objects to span the space of realistic images for training and evaluates the method on VIRAT, UA-DETRAC, EPFL-Car datasets, where it is demonstrated that using scene specific domain randomized synthetic data is better than fine-tuning off-the-shelf models on limited real data.
- An Independent and Interactive Museum Experience for Blind People53
A solution to support an independent, interactive museum experience that uses the continuous tracking of the user's location and orientation to enable a seamless interaction between Navigation and Art Appreciation.
- 53
- Environmental Factors in Indoor Navigation Based on Real-World Trajectories of Blind Users51
This work proposes two measures of path-following behavior: deviation from optimal route and trajectory variability and identifies relationships between these measures and elements of the environment, route characteristics, localization error, and instructional cues that users receive.
- Learning Action Maps of Large Environments via First-Person Vision49
This work demonstrates that by capturing appearance-based attributes of the environment and associating these attributes with activity demonstrations, the proposed mathematical framework allows for the prediction of Action Maps in new environments.
- Zero-Shot Video Question Answering with Procedural Programs46
This work proposes to answer zero-shot questions about videos by generating short procedural programs that derive a final answer from solving a sequence of visual subtasks, using a large language model to generate such programs from an input question and an API of visual modules in the prompt, then executes them to obtain the output.
- Importance of Self-Consistency in Active Learning for Semantic Segmentation46
The task of active learning in the context of semantic segmentation is addressed and it is shown that self-consistency can be a powerful source of self-supervision to greatly improve the performance of a data-driven model with access to only a small amount of labeled data.
- Neural Batch Sampling with Reinforcement Learning for Semi-supervised Anomaly Detection46
A novel semi-supervised learning algorithm for anomaly detection and segmentation using an anomaly class that uses as input the loss profile of a data sample processed through an autoencoder, and a Reinforcement Learning based meta-algorithm, which is called the neural batch sampler, to strategically sample training batches during autoencoder training.
- How Context and User Behavior Affect Indoor Navigation Assistance for Blind People46
This work performs a fine-grained analysis of user behavior during indoor assisted navigation, outlining different scenarios where user behavior (either with a white-cane or a guide-dog) is likely to cause navigation errors.
- SMPLOlympics: Sports Environments for Physically Simulated Humanoids44
This work presents SMPLOlympics, a collection of physically simulated environments that allow humanoids to compete in a variety of Olympic sports, and shows that combining strong motion priors with simple rewards can result in human-like behavior in various sports.
- 44
- Crowdsourcing the Installation and Maintenance of Indoor Localization Infrastructure to Support Blind Navigation43
LuzDeploy is a physical crowdsourcing system that organizes non-experts for the installation and long-term maintenance of a Bluetooth-based navigation system and aims to improve the sustainability of indoor navigation systems to encourage widespread adoption outside of research settings.
- Making GIFs Accessible42
It is recommended that social media platforms automatically include alt text descriptions for popular GIFs (as Twitter has begun to do), and content producers create audio descriptions to ensure everyone has a rich and emotive experience with GIFs online.
- MonoEye42
The MonoEye system is presented, a multimodal human motion capture system using a single RGB camera with an ultra-wide fisheye lens, mounted on the user's chest, that is possible to recognize various motions that existing egocentric motion capture systems cannot recognize.
- Joint Detection and Multi-Object Tracking with Graph Neural Networks41
This work proposes a new approach for joint MOT based on Graph Neural Networks (GNNs), which can explicitly model complex interactions between multiple objects in both the spatial and temporal domains, essential for learning discriminative features for detection and data association.
- Efficient Non-Line-of-Sight Imaging from Transient Sinograms41
This work explores a more efficient form of NLOS scanning that reduces both acquisition times and computational requirements and proposes a circular and confocal non-line-of-sight scan that involves illuminating and imaging a common point, and scanning this point in a circular path along a wall.
- Synthesizing a Scene-Specific Pedestrian Detector and Pose Estimator for Static Video Surveillance41
It is demonstrated that when real human annotated data is scarce or non-existent, the data generation strategy can provide an excellent solution for an array of tasks for human activity analysis including detection, pose estimation and segmentation.
- TEMPO: Efficient Multi-View Pose Estimation, Tracking, and Forecasting39
TEMPO is an efficient multi-view pose estimation model that learns a robust spatiotemporal representation, improving pose accuracy while also tracking and forecasting human pose, and is able to generalize across datasets without scene-specific fine-tuning.
- 39
- 39
- HOIGPT: Learning Long-Sequence Hand-Object Interaction with Language Models38
A novel physically grounded HOI tokenizer, the hand-object decomposed VQ-VAE, for discretizing HOI sequences, and a motion-aware language model trained to process and generate both text and HOI tokens are introduced.
- Joint Metrics Matter: A Better Standard for Trajectory Forecasting38
This work presents the first comprehensive evaluation of state-of-the-art (SOTA) trajectory forecasting methods with respect to multi-agent metrics (joint metrics): JADE, JFDE, and collision rate, and demonstrates the importance of joint metrics as opposed to marginal metrics.
- Impact of Expertise on Interaction Preferences for Navigation Assistance of Visually Impaired Individuals38
This study motivates the design of future navigation systems capable of verbosity level personalization in order to keep the users engaged in the current situational context while minimizing distractions.
- Forecasting Time-to-Collision from Monocular Video: Feasibility, Dataset, and Challenges38
A more fine-grained approach to collision forecasting is proposed by predicting the exact time to collision in terms of milliseconds, which is more helpful for collision avoidance in the context of dynamic path planning.
- Deformer: Dynamic Fusion Transformer for Robust Hand Pose Estimation37
The Deformer is proposed: a framework that implicitly reasons about the relationship between hand parts within the same image and different timesteps and is more robust to occlusions than state-of-the-art methods.
- Improving Lesion Segmentation for Diabetic Retinopathy Using Adversarial Learning37
An end-to-end system for pixel-level segmentation of DR lesions is proposed by incorporating HEDNet into a Conditional Generative Adversarial Network (cGAN), and a loss function that adds adversarial loss to segmentation loss is designed.
- S2Net: Stochastic Sequential Pointcloud Forecasting36
This work tackles the stochastic SPF problem by proposing a generative model with two main components: a conditional variational recurrent neural network that models a temporally-dependent latent space and a pyramid-LSTM that increases the fidelity of predictions with temporalally-aligned skip connections.
- 36
- 36
- 36
- 35
- Joint 3D Tracking and Forecasting with Graph Neural Network and Diversity Sampling35
This work proposes a unified solution for 3D MOT and trajectory forecasting which also incorporates two additional novel computational units and introduces Graph Neural Networks to capture the way in which multiple agents interact with one another.
- Generative Hybrid Representations for Activity Forecasting With No-Regret Learning34
This work develops an efficient deep generative model to jointly forecast a person's future discrete actions and continuous motions and proposes a variant to continually learn this model from streaming data, to observe its practical effectiveness and theoretically justify its learning efficiency.
- Efficient K-Shot Learning With Regularized Deep Networks34
This work proposes a simple yet effective regularization method for fine-tuning pre-trained deep networks for the task of k-shot learning and proposes a reinforcement learning search strategy using recurrent networks to learn the optimal group assignments for each network layer.
- Variability in Reactions to Instructional Guidance during Smartphone-Based Assisted Navigation of Blind Users32
A data-driven analysis of reaction variability as defined by motion and timing measures finds significant variability between users in their reaction characteristics during real-world navigation with a deployed system.
- 32
- 30
- Wide-Baseline Multi-Camera Calibration using Person Re-Identification30
This work proposes a method that uses ideas from person re-identification (re-ID) for wide-baseline camera calibration, which first employs a re-ID method to associate human bounding boxes across cameras, then converts bounding box correspondences to point correspondences, and finally solves for camera pose using multi-view geometry and bundle adjustment.
- 30
- VizMap30
The VizMap system uses computer vision and crowdsourcing to collect information and make it available non-visually and move towards integrating the strengths of the end user, on-site crowd, online crowd, and computer vision to solve a long-standing challenge in indoor blind exploration.
- Real-Time Simulated Avatar from Head-Mounted Sensors29
SimXR is presented, a method for controlling a simulated avatar from information (headset pose and cameras) ob-tained from AR / VR headsets that does not rely on any intermediate representations and learns to directly map from images and headset poses to humanoid control signals.
- Modeling Expertise in Assistive Navigation Interfaces for Blind People29
A step towards tailoring the navigation interface to the user»s needs is taken by proposing a personalized recurrent neural network-based behavior model for expertise level classification.
- 29
- GroundNet27
This model achieves the top-ranked performance on ground plane normal estimation and horizon line detection on the real-world outdoor datasets of ApolloScape and KITTI, improving the performance of previous art by up to 17.7% relatively.
- ExpertAF: Expert Actionable Feedback from Video26
This work introduces a novel method to generate actionable feedback (AF) from video of a person doing a physical activity, such as basketball or soccer, and devise a multimodal video-language model to infer coaching feedback.
- ZeroGrasp: Zero-Shot Shape Reconstruction Enabled Robotic Grasping26
This work introduces ZeroGrasp, a novel framework that simultaneously performs 3D reconstruction and grasp pose prediction in near real-time and generalizes to novel real-world objects by leveraging synthetic data.
- 26
- RECOVERING THE BASIC STRUCTURE OF HUMAN ACTIVITIES FROM NOISY VIDEO-BASED SYMBOL STRINGS26
This paper presents a framework for identifying noise and recovering the basic activity grammar from a noisy symbol string produced by video, and shows the robustness of the system to noise and its effectiveness in learning the basic structure of human activity.
- 25
- 25
- Smart Skin: Vision-Based Soft Pressure Sensing System for In-Home Hand Rehabilitation24
A novel in-home hand rehabilitation system for monitoring hand motions and assessing grip forces of stroke patients with minimal or no supervision by recording the training process and the force data, which can be automatically conveyed to physical therapists.
- 24
- 24
- Audio-Visual Self-Supervised Terrain Type Recognition for Ground Mobile Platforms22
This work proposes a multi-modal self-supervised learning technique that switches between audio features extracted from a microphone attached to the underside of a mobile platform and image features extracted by a camera on the platform to cluster terrain types, which achieves over 80% accuracy.
- 22
- 21
- Incremental Class Discovery for Semantic Segmentation With RGBD Sensing21
A novel method that incrementally learns new classes for image segmentation using RGBD sensing to discover new semantic classes over time and is able to correctly cluster objects of both known and unseen classes is proposed.
- Unlocking Exocentric Video-Language Data for Egocentric Video Representation Learning20
A data transformation framework to adapt exocentric data for egocentric training, focusing on identifying specific video clips that emphasize hand-object interactions and transforming narration styles to align with egocentric perspectives, which creates a new egocentric dataset derived from exocentric video-language data.
- 20
- Adversarial domain adaptation for cross data source macromolecule in situ structural classification in cellular electron cryo-tomograms20
A deep learning-based adversarial domain adaptation (3D-ADA) method is adapted to timely address the domain shift problem in CECT data analysis and can improve cross data source recovery of novel macromolecular structures.
- Long-Term Activity Forecasting Using First-Person Vision20
This work develops the first forecasting framework that uses ego-centric video input, which is an optimal vantage point for understanding human activities over large spaces and makes it possible to combine a high dimensional continuous state with a discrete state portion into a single state for behavior forecasting.
- 20
- 19
- 19
- Track Targets by Dense Spatio-Temporal Position Encoding18
The proposed paradigm, Dense Spatio-Temporal (DST) position encoding, encodes spatio-temporal position information in a pixel-wise dense fashion and provides location information to associate targets across frames beyond appearance matching by comparing objects in two bounding boxes.
- KDFNet: Learning Keypoint Distance Field for 6D Object Pose Estimation18
KDFNet is presented, a novel method for 6D object pose estimation from RGB images able to robustly estimate the 6D pose in challenging scenarios including occlusion and achieves state-of-the-art performance on Occlusion LINEMOD dataset.
- AEI: Actors-Environment Interaction with Adaptive Attention for Temporal Action Proposals Generation18
This paper makes an attempt to simulate that ability of a human by proposing Actor Environment Interaction (AEI) network to improve the video representation for temporal action proposals generation and temporal action detection.
- 18
- SolePoser: Full Body Pose Estimation using a Single Pair of Insole Sensor17
The proposed SolePoser system utilizes pressure and IMU sensors embedded in insoles to capture the body weight’s pressure distribution at the feet and its 6 DoF acceleration, which is used to estimate the 3D full-body joint position by a two-stream transformer network.
- 17
- 17
- 17
- Accelerating Vision Transformers with Adaptive Patch Sizes16
Adaptive Patch Transformers (APT) significantly reduces training and inference time without loss of performance in high-resolution dense visual tasks, achieving up to 30\% faster training and inference in visual QA, object detection, and semantic segmentation.
- From Universal Humanoid Control to Automatic Physically Valid Character Creation16
A humanoid design framework to automatically generate physically valid humanoid bodies conditioned on sequence(s) of pre-specified human motions and uses a design-and-control framework to optimize a humanoid’s physical attributes to find body designs that can better imitate the pre- specified human motion sequence.
- Sequential Voting with Relational Box Fields for Active Object Detection16
This work adopts standard RL to learn the voting function parameters and shows that it provides a meaningful improvement over a standard supervised learning approach, and performs experiments on two large-scale datasets: 100DOH and MECCANO, improving AP50 performance by 8% and 30% over the state of the art.
- Future Near-Collision Prediction from Monocular Video: Feasibility, Dataset, and Challenges.16
A more fine-grained approach to collision forecasting is proposed by predicting the exact time to collision in terms of milliseconds, which is more helpful for collision avoidance in the context of dynamic path planning.
- 16
- Human-centric panoramic imaging stitching16
An image mosaicing algorithm is presented that is robust to parallax and misalignment, and is also able to preserve the important human-centric content, specifically faces, and finds an optimal path between the boundary of two images that preserves color continuity and peoples' faces in the scene.
- Ego4D: Around the World in 3,600 Hours of Egocentric Video15
Ego4D, a massive-scale egocentric video dataset and benchmark suite, introduced, dramatically expands the volume of diverse egocentric video footage publicly available to the research community and presents a host of new benchmark challenges centered around understanding the first-person visual experience.
- 15
- Visual Motif Discovery via First-Person Vision15
A commonality clustering method is developed that leverages three important aspects: inter-video similarity, intra-video sparseness, and people’s visual attention to discover visual motifs from a collection of first-person videos recorded by a wearable camera.
- A Game-Theoretic Approach to Multi-Pedestrian Activity Forecasting15
Predictive models of pedestrian dynamics are developed by encoding the coupled nature of multi-pedestrian interaction using game theory, and deep learning-based visual analysis to estimate person-specific behavior parameters.
- BallCam!15
Despite the challenge of extreme camera motion, it is shown that it can leverage the periodicity of spinning cameras to generate a stabilized ball point-of-view video.
- Detecting Interesting Events Using Unsupervised Density Ratio Estimation15
This paper defines interesting events as unusual events which occur rarely in the entire video and proposes a novel interesting event summarization framework based on the technique of density ratio estimation recently introduced in machine learning.
- 3-D interaction with a large wall display using transparent markers15
A new interface for interacting with large displays via small video devices such as a cell phone using a matrix of transparent markers embedded on display that allows the user to interact with digital contents without being distracted by opaque visual markers.
- ATLAS: Decoupling Skeletal and Shape Parameters for Expressive Parametric Human Modeling14
ATLAS outperforms existing methods by fitting unseen subjects in diverse poses more accurately, and quantitative evaluations show that the non-linear pose correctives more effectively capture complex poses compared to linear models.
- 14
- Helping People Through Space and Time: Assistance as a Perspective on Human-Robot Interaction14
A definition of assistance is presented, a survey of existing work, and three key design axes that occur in many domains and benefit from the examination of assistance as a whole are proposed as useful tools for unifying assistance research across domains.
- Multi-Echo LiDAR for 3D Object Detection14
This work presents a 3D object detection model which leverages the full spectrum of measurement signals provided by LiDAR, and proposes a multi-signal fusion module to combine the reflectance and ambient features extracted with a 2D CNN, and point cloud features extracted using a3D graph neural network (GNN).
- When We First Met: Visual-Inertial Person Localization for Co-Robot Rendezvous14
This work proposes a method to learn a visual-inertial feature space in which the motion of a person in video can be easily matched to the motion measured by a wearable inertial measurement unit (IMU).
- Multiple Networks are More Efficient than One: Fast and Accurate Models via Ensembles and Cascades.14
This work demonstrates a simple complementary paradigm to obtain efficient and accurate models that requires no architectural tuning, and shows that committee-based models, i.e., ensembles or cascades of models, can easily obtain higher accuracy with less computation when compared to a single model.
- Zero-Shot Multi-object Scene Completion13
OctMAE, an architecture that leverages an Octree U-Net and a latent 3D MAE to achieve high-quality and near real-time multi-object scene completion through both local and global geometric reasoning, is proposed.
- 13
- 13
- 13
- Semi-automated home-based therapy for the upper extremity of stroke survivors13
This work presents a home-based system that integrates customized therapy objects, camera based movement capture and assessment techniques, and a flexible exercise protocol aimed at generalizing to variable daily life activities.
- 13
- VADRA: Visual Adversarial Domain Randomization and Augmentation.13
This work proposes a novel algorithm which closes the loop between the synthetic generative model and the learner in an adversarial fashion and evaluates its method on diverse vision tasks using state-of-the-art simulators for public datasets like CLEVR, Syn2Real, and VIRAT, to demonstrate that a learner trained using adversarial data generation performs better than using a random data generation strategy.
- 13
- 13
- Human Action Anticipation: A Survey12
This survey aims to tie together this fragmented literature on behavior prediction, covering recent technical innovations as well as the development of new large-scale datasets for model training and evaluation.
- 12
- 12
- Multi-Modality Task Cascade for 3D Object Detection12
A novel MultiModality Task Cascade network (MTC-RCNN) that leverages 3D box proposals to improve 2D segmentation predictions, which are then used to further refine the 3D boxes and shows that including a 2D network between two stages of 3D modules significantly improves both 2D and 3D task performance.
- Rotational Rectification Network: Enabling Pedestrian Detection for Mobile Vision12
A Rotational Rectification Network (R2N) that can be inserted into any CNNbased pedestrian (or object) detector to adapt it to significant changes in camera rotation is proposed and shown how it can improve the performance of state-of-the-art pedestrian detectors under heavy image rotation by up to 45%.
- 12
- Origami Sensei: A Mixed Reality AI-Assistant11
An AI-augmented Mixed Reality approach for learning hands-on creative tasks by introducing Origami Sensei as an approach for learning origami shows that participants prefer Origami Sensei, and it increases task efficiency and learner engagement.
- 11
- HERD: Continuous Human-to-Robot Evolution for Learning from Human Demonstration11
This work shows that manipulation skills can be transferred from a human to a robot through the use of micro-evolutionary reinforcement learning, where a five-finger human dexterous hand robot gradually evolves into a commercial robot, while repeated interacting in a physics simulator to continuously update the policy that is first learned from human demonstration.
- 11
- HOMER11
HOMER is an interactive system that uses novel therapy artifacts, a computer vision approach, and a tablet interface to provide users with a flexible solution suitable for home based rehabilitation.
- 11
- Sequential Forecasting of 100,000 Points10
This work studies the problem of future prediction at the level of 3D scenes, represented by point clouds captured by a LiDAR sensor, and presents a new object trajectory forecasting pipeline leveraging SPCSFNet, able to outperform the conventional pipeline using state-of-the-art trajectory forecasting methods.
- Emergent Active Perception and Dexterity of Simulated Humanoids from Visual Reinforcement Learning9
Perceptive Dexterous Control is introduced, a framework for vision-driven dexterous whole-body control with simulated humanoids that enables learning a single policy to perform multiple household tasks, including reaching, grasping, placing, and articulated object manipulation.
- Origami Sensei: Mixed Reality AI-Assistant for Creative Tasks Using Hands9
Origami Sensei is introduced, a mixed-reality system that assists beginners in creating origami by providing step-by-step instructions in real-time, and enables users to create their own origami pieces from a library of pre-existing designs.
- Unsupervised Sequence Forecasting of 100,000 Points for Unsupervised Trajectory Forecasting.9
This work studies the problem of future prediction at the level of 3D scenes, represented by point clouds captured by a LiDAR sensor, and presents a new object trajectory forecasting pipeline leveraging SPCSFNet, able to outperform the conventional pipeline using state-of-the-art trajectory forecasting methods.
- Phase-Parametric Policies for Reinforcement Learning in Cyclic Environments9
This work proposes phase-parameterized policies and value function approximators that explicitly enforce a cyclic structure to the policy or value space and applies this approach to both feed-forward and recurrent deep networks in the context of trajectory optimization and locomotion problems.
- 8
- JaywalkerVR: A VR System for Collecting Safety-Critical Pedestrian-Vehicle Interactions8
This work proposes a virtual reality human-in-the-loop simulator, JaywalkerVR, to obtain vehicle-pedestrian interaction data from safety critical scenarios and creates a high-quality dataset with vehicle-pedestrian interaction data from safety critical scenarios called CARLA-VR, which addresses the lack of long-tail data samples in commonly used real world autonomous driving datasets.
- 8
- Multi-Person 3D Pose Estimation from Multi-View Uncalibrated Depth Cameras8
This work presents a simple pipeline for Multi-View Depth Human Pose Estimation (MVD-HPE), which outperforms the current 3D human pose regression-free pipelines in terms of both camera pose estimation and 3D human pose estimation.
- Learnable Spatio-Temporal Map Embeddings for Deep Inertial Localization8
A data-driven prior that learns to encode which map regions are feasible locations for a user more accurately than previous hand-defined methods is proposed, which leads to a 49% improvement in inertial-only localization accuracy when used in a particle filter.
- Inverse Reinforcement Learning with Explicit Policy Estimates8
This work identifies various problem scenarios and investigates each method's suitability for these problems, and demonstrates key computational and algorithmic differences which arise between the methods due to an approximation of the optimal soft value function.
- 8
- 8
- NovaPlan: Zero-Shot Long-Horizon Manipulation via Closed-Loop Video Language Planning7
NovaPlan is introduced, a hierarchical framework that enables robust, zero-shot long-horizon manipulation by systematically proposing, verifying, and repairing visual plans, and utilizes a hybrid geometric representation that adaptively switches between object-centric flow and human hand flow.
- Towards Online Adaptation for Autonomous Household Assistants7
It is shown, under certain assumptions, that representing objects by their material properties can increase sample efficiency and generalizability to out of domain objects.
- 7
- Neighborhood-Aware Neural Architecture Search7
A novel neighborhood-aware NAS formulation to identify flat-minima architectures in the search space, with the assumption that flat minima generalize better than sharp minima, which achieves state-of-the-art performance on established benchmarks.
- Estimating 3D Camera Pose from 2D Pedestrian Trajectories7
A data-driven approach by training a neural network (NN) regressor to model a direct mapping from 2D pedestrian trajectories projected on the image plane to 3D camera pose, which achieves an improvement of ~ 50% on both position and orientation prediction accuracy when compared to other SOTA methods.
- 7
- Experiencing the ball's POV for ballistic sports7
A robust video synthesis algorithm is proposed that leverages the unique constraints of fast spinning cameras to obtain a stabilized bird's eye point-of-view video clip from a small wireless camera inside an American football.
- DegustaBot: Zero-Shot Visual Preference Estimation for Personalized Multi-Object Rearrangement6
DegustaBot is presented, an algorithm for visual preference learning that solves household multi-object rearrangement tasks according to personal preference using internet-scale pre-trained vision-and-language foundation models with novel zero-shot visual prompting techniques.
- 3D-Aware Encoding for Style-based Neural Radiance Fields6
A two-stage encoder is proposed for style-based NeRF inversion that qualitatively and quantitatively exhibits superiority over the existing encoders for inversion in both image reconstruction and novel-view rendering.
- 6
- 6
- GroundNet: Segmentation-Aware Monocular Ground Plane Estimation with Geometric Consistency.6
This work focuses on the problem of estimating the orientation of the ground plane with respect to a mobile monocular camera platform by jointly optimizing for point-wise surface normal direction, 2D ground segmentation, and depth estimates.
- 6
- Online Semantic Activity Forecasting with DARKO.6
A framework that continuously forecasts long-term spatial and semantic intentions (what you will do and where you will go) of a first-person camera wearer is presented, and the algorithm is named “Demonstrating Agent Rewards for K-futures Online” (DARKO).
- Hybrid macro–micro visual analysis for city-scale state estimation6
This work addresses the task of estimating large-scale land surface conditions using overhead aerial and street view images and micro-level images, and uses macro-level imaging to learn the extent to which the land condition corresponds between land regions that share similar visual characteristics.
- CacheFlow: Fast Human Motion Prediction by Cached Normalizing Flow5
This work introduces a novel flow-based method for human motion prediction called CacheFlow, which takes advantage of an unconditional flow-based generative model that transforms a Gaussian mixture into the density of future motions.
- Multi-Object Tracking by Hierarchical Visual Representations5
A new visual hierarchical representation paradigm for multi-object tracking that achieves state-of-the-art accuracy and time efficiency among query-based methods on multiple multi-object tracking benchmarks is proposed.
- Multi-View Person Matching and 3D Pose Estimation with Arbitrary Uncalibrated Camera Networks5
The PME method outperforms other methods by a large margin on cross-view person matching, reaches SOTA performance on 3D human pose estimation without using either camera poses or 3D training data, and shows good generalization ability across five datasets of various environment settings.
- 5
- MGpi: A Computational Model of Multiagent Group Perception and Interaction5
The efficacy of the KPM gate as a social attention mechanism is demonstrated, achieving state-of-the-art performance on the task of group identification without using explicit group annotations, layout assumptions, or manually chosen parameters.
- 5
- Massive City-Scale Surface Condition Analysis Using Ground and Aerial Imagery5
This work presents a unified framework for robustly integrating image data taken at vastly different viewpoints to generate large-scale estimates of land surface conditions and evaluates the approach on two modes of land condition analysis, namely, city-scale debris and greenery estimation.
- Discovering Primitive Action Categories by Leveraging Relevant Visual Context5
This work learns a bi-modal latent variable model to discover primitive action categories from video without supervision and shows that the combination of relevant visual context and motion features improves the performance of action discovery.
- 4
- 4
- 4
- 4
- 4
- 4
- Adversarially Optimizing Intersection over Union for Object Localization Tasks4
This work addresses uncertainty adversarially by approximating provided datasets annotations within a game-theoretic formulation of prediction tasks, constrained to resemble the training data annotations according to a set of specified features.
- Generalizable Neural Human Renderer3
The challenging task of learning a Generalizable Neural Human Renderer (GNH) is tackled, a novel method for rendering animatable humans from monocular video without any test-time optimization, notably achieving a 31.3% improvement in LPIPS.
- 3
- 3
- Cost-Aware Evaluation and Model Scaling for LiDAR-Based 3D Object Detection3
A cost-aware evaluation of LiDAR-based 3D object detectors focuses on SECOND, a simple grid-based one-stage detector, and finds that, if allowed to use the same latency, SECOND can match the performance of PV-RCNN++, the current state-of-the-art method on the Waymo Open Dataset.
- 3
- 3
- 3
- 3
- Learning Context-dependent Personal Preferences for Adaptive Recommendation3
It is shown that the proposed algorithms exhibit better prediction accuracy compared to traditional supervised learning and bandit algorithms, and are robust under challenging limited prediction settings in which a smaller number of expert policies is assumed.
- Personalized Dynamics Models for Adaptive Assistive Navigation Interfaces.3
PING (Personalized INstruction Generation agent), a model-based reinforcement learning framework which aims to quickly adapt its state transition dynamics model to match the reactions of the user using a novel end-to-end learned weighted majority-based regression algorithm, is proposed.
- Activity Forecasting3
A case is made for a decision-theoretic approach to human activity forecasting, which provides a principled framework for modeling the consequences of taking certain actions and the impact it can have on the future.
- 3
- 3
- Automating Stroke Rehabilitation for Home-Based Therapy3
A conceptual design to automatically evaluate a subject’s performance for a homebased stroke rehabilitation system is presented and it is proposed to model a reaching task as a trajectory in the state space of hand part features and then use reward learning to automatically generate new ratings for subjects to track performance over time.
- 2
- S2GO: Streaming Sparse Gaussian Occupancy Prediction2
Owing to its efficient, query-based representation, S2GO achieves state-of-the-art performance on the nuScenes and KITTI occupancy benchmarks, outperforming prior art (e.g., GaussianWorld) by 1.5 IoU with 5.9x faster inference.
- 2
- Evaluating a VR System for Collecting Safety-Critical Vehicle-Pedestrian Interactions2
It is demonstrated that the VR-based data collection system elicits realistic responses for capturing pedestrian data in safety-critical or uncommon vehicle-pedestrian interaction scenarios.
- 2
- 2
- Modeling Social Group Communication with Multi-Agent Imitation Learning.2
The KPM-Gate can be used to discover social groups using its natural interpretation as a social attention mechanism and experimental validation shows that the model is able to both discover social group structure and learn an accurate multi-agent communication policy.
- 2
- 2
- 2
- Recognizing Multiple Objects via Regression Incorporating the Co-occurrence of Categories2
This paper presents a novel method for object recognition that explicitly deals with objects of multiple categories coexisting in an image through MAP regression, and aims to recognize objects by taking advantage of a scene's context represented by the co-occurrence relationship between object categories.
- RECOGNIZING OVERLAPPED HUMAN ACTIVITIES FROM A SEQUENCE OF PRIMITIVE ACTIONS VIA DELETED INTERPOLATION2
A hierarchical Bayesian network based on a stochastic context-free grammar is implemented to address the hierarchical nature of human activity recognition and results from the analysis of action sequences based on video surveillance data show the validity of the approach.
- 2
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- Video segmentation and stabilization for BallCam1
An activity-aware video stabilization algorithm is proposed which is able to understand the current activity of the BallCam, which uses estimated activity labels to inform a robust video stabilized algorithm.
- Rotational Rectification Network for Robust Pedestrian Detection1
A rotational rectification network that can be inserted into any CNN-based pedestrian (or object) detector to maintain performance despite significant rotational variance and is used to enhance the performance of a state-of-the-art pedestrian detector under heavy image rotation.
- Recognizing multiple objects based on co-occurrence of categories1
This paper presents a novel method for object recognition that explicitly deals with objects of multiple categories coexisting in an image through MAP regression, and aims to recognize objects by taking advantage of a scene’s context represented by the co-occurrence relationship between object categories.
- Online No-regret Model-Based Meta RL for Personalized Navigation–
This work proposes an online no-regret model-based RL method that quickly conforms to the dynamics of the current user and can reduce the number of collisions by more than 60%.
- –
- –
- –
- –
- –
- –
- OMG: Orthogonal Method of Grouping With Application of K-Shot Learning–
This paper proposes a generalizable k-shot learning framework that can be used on any pre-trained network, by grouping network parameters to produce a low-dimensional representation of the parameter space.
- Activity-Aware Video Stabilization for BallCam–
An activity-aware video stabilization algorithm is proposed which is able to understand the current activity of the BallCam, which uses estimated activity labels to inform a robust video stabilization algorithm.
- –
- Semantic video segmentation using both appearance and geometric information–
The results show that the proposed algorithm outperforms other appearance based segmentation method in terms of semantic quality of the segmentation and will serve as a basis for better high-level tasks such as recognition, tracking, and video understanding.
- Dynamic View Synthesis from Camera in Spinning Ball–
A robust video synthesis algorithm is proposed that leverages the unique constraints of fast spinning cameras to obtain a stable bird’s-eye-view video clip and uses coarse-to-fine image homography computation to progressively register images.
- Ego-motion analysis using average image data intensity–
A new method to perform ego-motion analysis using intensity averaging of image data that can estimate general motions from two sequential images on pixel plane by calculating cross correlations and can operate at up to 714 FPS in calculating one dimensional translation motion.
- –
- –
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 2026-10-14. 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.