Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Pascal VasseurSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works206 from public data
- Transforming Model Prediction for Tracking433
This work trains the proposed tracker end-to-end and validate its performance by conducting comprehensive experiments on multiple tracking datasets, achieving an AUC of 68.5% on the challenging LaSOT [14] dataset.
- Learning Target Candidate Association to Keep Track of What Not to Track388
This work proposes a training strategy that combines partial annotations with self-supervision to keep track of distractor objects in order to continue tracking the target, and introduces a learned association network to propagate the identities of all target candidates from frame-to-frame.
- Covariance Pooling for Facial Expression Recognition181
The benefits of using a manifold network structure for covariance pooling to improve facial expression recognition are explored and the advantage of pooling image-set features temporally by stacking the designed manifold network of covariance Pooling on top of convolutional network layers is demonstrated.
- Single-Model and Any-Modality for Video Object Tracking172
This work introduces Un-Track, a Unified Tracker of a single set of parameters for any modality, which surpasses both SOTA unified trackers and modality-specific counterparts, validating the effectiveness and practicality.
- Structured Bird’s-Eye-View Traffic Scene Understanding from Onboard Images158
This work studies the problem of extracting a directed graph representing the local road network in BEV coordinates, from a single onboard camera image, and shows that the method can be extended to detect dynamic objects on the BEV plane.
- Sliced Wasserstein Generative Models150
This paper proposes to approximate SWDs with a small number of parameterized orthogonal projections in an end-to-end deep learning fashion and designs two types of differentiable SWD blocks to equip modern generative frameworks---Auto-Encoders and Generative Adversarial Networks.
- Source-free Depth for Object Pop-out142
This work adapts depth inference models for object segmentation using the objects’ "pop-out" prior in 3D to adapt the inferred depth maps such that objects can be localized using only 3D information.
- 130
- Complexity Experts are Task-Discriminative Learners for Any Image Restoration128
The proposed MoCE-IR model outperforms state-of-the-art methods, affirming its efficiency and practical applicability and introducing “complexity experts” – flexible expert blocks with varying computational complexity and receptive fields.
- The Ninth Visual Object Tracking VOT2021 Challenge Results127
The Visual Object Tracking challenge VOT2021 is the ninth annual tracker benchmarking activity organized by the VOT initiative; results of 71 trackers are presented; many are state-of-the-art trackers published at major computer vision conferences or in journals in recent years.
- Unsupervised Domain Adaptation for Nighttime Aerial Tracking125
A unique object discovery approach is provided to generate training patches from raw nighttime tracking videos to tackle the domain discrepancy and demonstrate the robustness and domain adaptability of the proposed framework in nighttime aerial tracking.
- A Continual Deepfake Detection Benchmark: Dataset, Methods, and Essentials91
A continual deepfake detection benchmark (CDDB) over a new collection of deepfakes from both known and unknown generative models is suggested, which is clearly more challenging than the existing benchmarks, and offers a suitable evaluation avenue to the future research.
- Learning to Relate Depth and Semantics for Unsupervised Domain Adaptation66
An approach for encoding visual task relationships to improve model performance in an Unsupervised Domain Adaptation (UDA) setting and an Iterative Self-Learning (ISL) training scheme, which exploits semantic pseudo-labels to provide extra supervision on the target domain.
- Topology Preserving Local Road Network Estimation from Single Onboard Camera Image60
This paper aims at extracting the local road network topology, directly in the bird’ s-eye- view (BEV), all in a complex urban set-ting, using a set of directed lane curves and their interactions, which are captured using their intersection points.
- SceneSplat: Gaussian Splatting-Based Scene Understanding with Vision-Language Pretraining59
This work introduces SceneSplat, to their knowledge the first large-scale 3D indoor scene understanding approach that operates natively on 3DGS, and proposes a self-supervised learning scheme that unlocks rich 3D feature learning from unlabeled scenes.
- 59
- Understanding Bird’s-Eye View of Road Semantics Using an Onboard Camera54
This work studies scene understanding in the form of online estimation of semantic BEV maps using the video input from a single onboard camera and proposes a novel architecture that combines these three aspects, which significantly surpasses the current state-of-the-art.
- Deformable Neural Radiance Fields using RGB and Event Cameras51
A novel method to model the deformable neural radiance fields using RGB and event cameras using RGB and event cameras using the asynchronous stream of events and calibrated sparse RGB frames is developed.
- 48
- Robust RGB-D Fusion for Saliency Detection48
This work proposes a robust RGB-D fusion method that benefits from layer-wise, and trident spatial, attention mechanisms, and demonstrates that the proposed fusion method performs consistently better than the state-of-the-art fusion alternatives.
- Domain Agnostic Feature Learning for Image and Video Based Face Anti-spoofing45
This paper proposes a class-conditional domain discriminator module, that, coupled with a gradient reversal layer, tries to generate live and spoof features that are discriminative, but at the same time robust against the aforementioned variability factors.
- 44
- Unsupervised Learning of Category-Specific Symmetric 3D Keypoints from Point Sets43
This paper aims at learning category-specific 3D keypoints, in an unsupervised manner, using a collection of misaligned 3D point clouds of objects from an unknown category, using the symmetric linear basis shapes without assuming the plane of symmetry to be known.
- 43
- Soft Contrastive Learning for Visual Localization41
This paper argues that any artificial division based on proximity measure is undesirable, due to the inherently ambiguous supervision for images near proximity threshold, and proposes a novel technique that uses soft positive/negative assignments of images for contrastive learning, avoiding the aforementioned problem.
- XTrack: Multimodal Training Boosts RGB-X Video Object Trackers39
This work proposes a mixture-of-experts framework designed for multimodal video object tracking, and implements a classifier with weak loss tasked with distinguishing between modalities, which shows the benefit of the proposed method for RGB-X tracker during inference, with an average precision improvement over the current SOTA.
- Cluster, Split, Fuse, and Update: Meta-Learning for Open Compound Domain Adaptive Semantic Segmentation39
This work proposes a principled meta-learning based approach to OCDA for semantic segmentation, MOCDA, by modeling the unlabeled target domain continuously, and achieves the state-of-the-art performance in both compound and open domains.
- 38
- 38
- ReVLA: Reverting Visual Domain Limitation of Robotic Foundation Models36
This work studies the visual generalization capabilities of three existing robotic foundation models, and proposes a gradual backbone reversal approach founded on model merging that enables OpenVLA - which requires the adaptation of the visual backbones during initial training - to regain its visual generalization ability.
- ObjectRelator: Enabling Cross-View Object Relation Understanding Across Ego-Centric and Exo-Centric Perspectives34
ObjectRelator is proposed, a novel approach featuring two key modules: Multimodal Condition Fusion (MCFuse) and SSL-based Cross-View Object Alignment (XObjAlign), which enforces cross-view consistency through self-supervised alignment, enhancing robustness to object appearance variations.
- Unsupervised Robust Domain Adaptation without Source Data33
The proposed method of using non-robust pseudo-labels performs surprisingly well on both clean and adversarial samples, for the task of image classification and shows a consistent performance improvement of over 10% in accuracy against the tested baselines on four benchmark datasets.
- 29
- Benchmarking Multi-Modal Semantic Segmentation Under Sensor Failures: Missing and Noisy Modality Robustness29
This work provides the first dedicated benchmark for MMSS robustness, offering new insights and tools to advance the field, and introduces a robustness benchmark that evaluates MMSS models under three scenarios: Entire-Missing Modality, Random-Missing Modality, and Noisy Modality.
- 29
- Efficient Conditional GAN Transfer with Knowledge Propagation across Classes29
A new GAN transfer method to explicitly propagate the knowledge from the old classes to the new classes, with the BN parameters increasing linearly with the number of new classes.
- 29
- NTIRE 2025 Challenge on Event-Based Image Deblurring: Methods and Results28
This paper presents an overview of NTIRE 2025, the First Challenge on Event-Based Image Deblurring, detailing the proposed methodologies and corresponding results, and anticipate that this challenge will drive further advancements in event-based vision research.
- Rethinking Global Context in Crowd Counting27
Extensive experiments on various datasets, including ShanghaiTech, UCF-QNRF, JHU-CROWD++ and NWPU, demonstrate that the proposed context extraction techniques can significantly improve the performance over the baselines.
- ShapeSplat: A Large-scale Dataset of Gaussian Splats and Their Self-Supervised Pretraining27
This work builds ShapeSplat, a large-scale dataset of 3DGS using the commonly used ShapeNet, ModelNet and Objaverse datasets, and introduces Gaussian-MAE, which highlights the unique benefits of representation learning from Gaussian parameters.
- 26
- 23
- What You Have is What You Track: Adaptive and Robust Multimodal Tracking22
This paper proposes a flexible framework for robust multimodal tracking through a novel Heterogeneous Mixture-of-Experts fusion mechanism with adaptive complexity, coupled with a video-level masking strategy that ensures both temporal consistency and spatial completeness - critical for effective video tracking.
- 21
- 21
- EgoNight: Towards Egocentric Vision Understanding at Night with a Challenging Benchmark21
EgoNight, the first comprehensive benchmark for nighttime egocentric vision, with visual question answering (VQA) as the core task, is presented, and EgoNight-VQA provides a strong foundation for advancing application-driven egocentric vision research and for developing models that generalize across illumination domains.
- Analogical Image Translation for Fog Generation21
An analogical image translation (AIT) that exploit the concept of gist, for the first time, to circumvent the challenge of collecting real foggy images, and achieves zero-shot image translation capability.
- Lego: Learning to Disentangle and Invert Personalized Concepts Beyond Object Appearance in Text-to-Image Diffusion Models20
Lego is introduced, a textual inversion method designed to invert subject-entangled concepts from a few example images that disentangles concepts from their associated subjects using a simple yet effective Subject Separation step and employs a Context Loss that guides the inversion of single/multi-embedding concepts.
- Low-Light Image Enhancement Using Event-Based Illumination Estimation19
The devised pipeline, dubbed Retinev, excels in producing well-illuminated, high dynamic range images, outperforming previous state-of-the-art event-based methods by up to $6.62 d B, while maintaining an efficient inference speed of 35.6 frames-per-second on a $640 \times 480$ image.
- 19
- GeoVLM-R1: Reinforcement Fine-Tuning for Improved Remote Sensing Reasoning17
A novel post training framework is proposed that incorporates task aware rewards to enable effective adaptation of reasoning based RL models to diverse EO tasks and enhances reasoning capabilities for remote sensing images, stabilizes optimization, and improves robustness.
- Incremental Non-Rigid Structure-from-Motion with Unknown Focal Length17
This paper presents a method for incremental Non-Rigid Structure-from-Motion (NRSfM) with the perspective camera model and the isometric surface prior with unknown focal length, and proposes a method to simultaneously recover the focal length and the non-rigid shapes.
- 16
- 16
- V$^{2}$-SAM: Marrying SAM2 with Multi-Prompt Experts for Cross-View Object Correspondence15
V2-SAM is presented, a unified cross-view object correspondence framework that adapts SAM2 from single-view segmentation to cross-view correspondence through two complementary prompt generators and introduces a Post-hoc Cyclic Consistency Selector (PCCS) that adaptively selects the most reliable expert based on cyclic consistency.
- Occam's LGS: An Efficient Approach for Language Gaussian Splatting15
This work shows that the sophisticated techniques for language-grounded 3D Gaussian Splatting are simply unnecessary and applies Occam’s razor to the task at hand and performs weighted multi-view feature aggregation using the weights derived from the standard rendering process, followed by a simple heuristic-based noisy Gaussian filtration.
- Facial Emotion Recognition with Noisy Multi-task Annotations15
This work introduces a new problem of facial emotion recognition with noisy multi-task annotations, and suggests a formulation from the point of joint distribution match view, which aims at learning more reliable correlations among raw facial images and multi- task labels, resulting in the reduction of noise influence.
- Unsupervised Compound Domain Adaptation for Face Anti-Spoofing15
A memory augmentation method for adapting the source model to the target domain in a domain aware manner, inspired by [29], which successfully adapts to the compound target domain consisting multiple new spoof types.
- Exo2EgoSyn: Unlocking Foundation Video Generation Models for Exocentric-to-Egocentric Video Synthesis14
Exo2EgoSyn is introduced, an adaptation of WAN 2.2 that unlocks Exocentric-to-Egocentric(Exo2Ego) cross-view video synthesis and enables high-fidelity ego view video generation from third-person observations without retraining from scratch.
- Convex Relaxations for Consensus and Non-Minimal Problems in 3D Vision14
The main contribution of this paper is the claim that a good approximate solution for many polynomial problems involved in 3D vision can be obtained using the existing theory of numerical computational algebra.
- Towards an Understanding of Our World by GANing Videos in the Wild.14
This work presents the first video generation framework that works in the wild, without making any assumption on the videos' content, and outperforms the state-of-the-art methods even when the static background assumption is valid.
- 14
- EarthMind: Leveraging Cross-Sensor Data for Advanced Earth Observation Interpretation with a Unified Multimodal LLM13
This work presents EarthMind, a novel vision-language framework for multi-granular and multi-sensor EO data understanding and outperforms existing methods on multiple public EO benchmarks, showcasing its potential to handle both multi-granular and multi-sensor challenges in a unified framework.
- Reducing Unimodal Bias in Multi-Modal Semantic Segmentation With Multi-Scale Functional Entropy Regularization13
A simple but effective plug-and-play regularization term based on functional entropy is applied, designed to intuitively balance the contribution of each visual modality to the segmentation results, which mitigates unimodal dominance and establishes a more balanced and robust segmentation framework.
- Articulate3D: Holistic Understanding of 3D Scenes as Universal Scene Description13
The value of Articulate3D is highlighted through cross-dataset and crossdomain evaluations and its applicability in downstream tasks such as scene editing through LLM prompting and robotic policy training for articulated object manipulation is showcased.
- Implicit Zoo: A Large-Scale Dataset of Neural Implicit Functions for 2D Images and 3D Scenes13
A large-scale dataset requiring thousands of GPU training days designed to facilitate research and development in implicit functions, which leads to an improved performance in all three task of image classification, semantic segmentation, and 3D pose regression, thereby unlocking new avenues for research.
- Event-Free Moving Object Segmentation from Moving Ego Vehicle13
This work introduces a novel large-scale dataset called DSEC-MOS for moving object segmentation from moving ego vehicles, which is the first of its kind, and devise EmoFormer, a novel network able to exploit the event data, which relies only on event data for training but does not require event input during inference.
- Geometrically Mappable Image Features13
This work proposes a novel loss function for learning image features which are both visually representative and geometrically relatable and shows that these features not only offer significantly better localization accuracy, but also allow to estimate the trajectory of a query sequence in absence of the reference images.
- 13
- PANORAMA: The Rise of Omnidirectional Vision in the Embodied AI Era12
An ideal panoramic system architecture in the embodied AI era, PANORAMA, is proposed, which consists of four key subsystems and highlights recent breakthroughs in omnidirectional generation, omnidirectional perception, omnidirectional understanding, and related datasets.
- 12
- 12
- A Unified and Interpretable Emotion Representation and Expression Generation12
This work proposes an in-terpretable and unified emotion model, referred as C2A2, and develops a method that leverages labels of the non-unified models to annotate the novel unified one and modify the text-conditional diffusion models to under-stand continuous numbers.
- Efficient Degradation-aware Any Image Restoration12
Extensive comparisons demonstrate that the DaAIR outperforms both state-of-the-art All-in-One models and degradation-specific counterparts, affirming the efficacy and practicality of this efficient restorer integrating holistic and specific learning within a unified model.
- Scaling Semantic Segmentation Beyond 1K Classes on a Single GPU12
This paper proposes a novel training methodology to train and scale the existing semantic segmentation models for a large number of semantic classes without increasing the memory overhead, and proposes an approximation method for ground-truth class probability, and uses it to compute cross-entropy loss.
- Improving Online Lane Graph Extraction by Object-Lane Clustering11
This work proposes an architecture and loss formulation to improve the accuracy of local lane graph estimates by using 3D object detection outputs to improve the accuracy of local lane graph estimates by using 3D object detection outputs.
- Image-based Navigation using Visual Features and Map.11
This work summarizes the reference images as a set of landmarks, which meet the requirements for image-based navigation, and formulates the map construction and self-localization problems as convex quadratic and second-order cone programs, respectively.
- 11
- 11
- 11
- AR-VLA: True Autoregressive Action Expert for Vision-Language-Action Models10
This work introduces a scalable, context-aware action generation schema that provides a robust structural foundation for training effective robotic policies and can effectively replace traditional chunk-based action heads for both specialist and generalist policies.
- Chorus: Multi-Teacher Pretraining for Holistic 3D Gaussian Scene Encoding10
Chorus is a multi-teacher pretraining framework that learns a holistic feed-forward 3D Gaussian Splatting scene encoder by distilling complementary signals from 2D foundation models, encouraging a shared embedding space that captures signals from high-level semantics to fine-grained structure.
- An Efficient Recurrent Adversarial Framework for Unsupervised Real-Time Video Enhancement10
This work proposes an efficient adversarial video enhancement framework that learns directly from unpaired video examples and introduces new recurrent cells that consist of interleaved local and global modules for implicit integration of spatial and temporal information.
- TACS: Taxonomy Adaptive Cross-Domain Semantic Segmentation10
This work addresses the image-level domain gap by proposing an uncertainty-rectified contrastive learning method, leading to more domain-invariant and class-discriminative features, and outperforms the previous state-of-the-art by a large margin.
- EgoSound: Benchmarking Sound Understanding in Egocentric Videos9
EgoSound is introduced, the first benchmark designed to systematically evaluate egocentric sound understanding in MLLMs, and establishes a challenging foundation for advancing multisensory egocentric intelligence, bridging the gap between seeing and truly hearing the world.
- Exploration-Driven Generative Interactive Environments9
This work proposes AutoExplore Agent - an exploration agent that entirely relies on the uncertainty of the world model, delivering diverse data from which it can learn the best, and can quickly adapt to new environments achieving video fidelity and controllability improvement.
- Diffusion-Based Particle-DETR for BEV Perception9
This work presents a diffusion-based DETR model for object detection that bears similarities to particle methods and analyzes the unique challenges of this approach, and presents a simple technique based on object query interpolation that allows the model to learn positional dependencies even in the presence of the diffusion noise.
- 9
- IMPACT: A Dataset for Multi-Granularity Human Procedural Action Understanding in Industrial Assembly8
IMPACT is the first real industrial assembly benchmark that jointly provides synchronized ego-exo RGB-D capture, decoupled bimanual annotation, compliance-aware state tracking, and explicit anomaly--recovery supervision within a single real industrial workflow.
- NeRF-GAN Distillation for Efficient 3D-Aware Generation with Convolutions8
This study aims at revisiting pose-conditioned 2D GANs for efficient 3D-aware generation at inference time by distilling 3D knowledge from pretrained NeRF-GANs and obtains results comparable with volumetric rendering in terms of quality and 3D consistency while benefiting from the computational advantage of convolutional networks.
- Sampling Algebraic Varieties for Robust Camera Autocalibration8
This paper addresses the problem of robustly autocalibrating a moving camera with constant intrinsics by using the Branch-and-Bound (BnB) search paradigm to maximize the consensus of the polynomials.
- 8
- VarCity - the Video8
VarCity - the Video is a short documentary-style CGI movie explaining the main outcomes of the 5-year Computer Vision research project VarCity, driven by some ad-hoc technical developments but more importantly of detailed and abundant communication and agreement on common best practices.
- RoHOI: Robustness Benchmark for Human-Object Interaction Detection7
The first robustness benchmark for HOI detection is introduced, evaluating model resilience under diverse challenges, and a Semantic-Aware Masking-based Progressive Learning (SAMPL) strategy is proposed to guide the model to be optimized based on holistic and partial cues, thus dynamically adjusting the model's optimization to enhance robust feature learning.
- CompositeTasking: Understanding Images by Spatial Composition of Tasks7
A novel task conditioning model is introduced – a single encoder-decoder network that performs multiple, spatially varying tasks at once that not only offers a compact network for multitasking, but also allows for task-editing.
- 7
- CityLoc: 6DoF Pose Distributional Localization for Text Descriptions in Large-Scale Scenes with Gaussian Representation6
This work introduces a method to generate distributions of camera poses conditioned on textual descriptions, facilitating robust reasoning for broadly defined concepts, and validates its superiority by comparing it against standard distribution estimation methods across five large-scale datasets.
- RICO: Two Realistic Benchmarks and an In-Depth Analysis for Incremental Learning in Object Detection6
Two Realistic Incremental Object Detection Benchmarks (RICO) are introduced, showing that all IL methods underperform in adaptability and retention, while replaying a small amount of previous data already outperforms all methods and individual training on the data remains superior.
- ConceptPose: Training-Free Zero-Shot Object Pose Estimation using Concept Vectors6
ConceptPose is introduced, a framework for object pose estimation that is both training-free and model-free and achieves state-of-the-art results on common zero shot relative pose estimation benchmarks, outperforming the strongest baseline by a relative 62\% in average ADD(-S).
- RefAtomNet++: Advancing Referring Atomic Video Action Recognition using Multi-Trajectory Semantic Retrieval6
RefAtomNet++ is introduced, a novel framework that advances cross-modal token aggregation through a multi-hierarchical semantic-aligned cross-attention mechanism combined with multi-trajectory Mamba modeling at the partial-keyword, scene-attribute, and holistic-sentence levels.
- Continuous Pose for Monocular Cameras in Neural Implicit Representation6
Using the assumption of continuous motion, changes in pose may actually live in a manifold that has lower than 6 degrees of freedom (DOF) is realized and the approach is used in vSLAM settings, showing impressive camera tracking performance.
- 2D–3D synchronous/asynchronous camera fusion for visual odometry6
A robust and direct 2D–3D registration method for camera synchronization and a visual odometry framework that benefits from both 2D and 3D acquisitions is proposed.
- Cross-View Multi-Modal Segmentation @ Ego-Exo4D Challenges 20255
A cross-view multi-modal object segmentation approach that enhances object localization by leveraging both visual masks and textual descriptions as segmentation conditions and introduces a cross-view object alignment module that enforces object-level consistency across perspectives, thereby improving the model's robustness to viewpoint changes.
- InTraGen: Trajectory-controlled Video Generation for Object Interactions5
This work introduces InTraGen, a pipeline for improved trajectory-based generation of object interaction scenarios, and introduces a multi-modal interaction encoding pipeline with an object ID injection mechanism that enriches object-environment interactions.
- 5
- Model-aware 3D Eye Gaze from Weak and Few-shot Supervisions5
The proposed method combines the best of both worlds by leveraging large amounts of weak annotations–which are easy to obtain, and only a few 3D gaze vectors–which alleviate the difficulty of fitting 3D eye models on the semantic segmentation of eye images.
- Online Lane Graph Extraction from Onboard Video5
This work uses the video stream from an onboard camera for online extraction of the surrounding's lane graph using the spatialtemporal embeddings in the transformer to enable the network to discover the best temporal aggregation strategy.
- Weakly Paired Multi-Domain Image Translation5
The proposed multi-domain image translation method consists of an attention-based generator and a two-stream discriminator that deals with misalignment between source and target images, including foreground objects such as cars and pedestrians.
- RobustFormer: Noise-Robust Pre-training for Images and Videos4
This work introduces RobustFormer, a novel framework that enables noise-robust masked autoencoder pre-training for both images and videos by using DWT for efficient downsampling, eliminating the need for expensive IDWT reconstruction and simplifying the attention mechanism to focus on noise-resilient multi-scale representations.
- Seeing Together: Multi-Robot Cooperative Egocentric Spatial Reasoning with Multimodal Large Language Models4
SP-CoR (Spectral and Physics-Informed Cooperative Reasoner), an MLLM framework for fine-grained cooperative spatial reasoning, is proposed, enabling the model to benefit from privileged robot-pose supervision during training while requiring only egocentric videos at test time.
- Understanding Museum Exhibits using Vision-Language Reasoning4
It is demonstrated the necessity of finetuning models on large-scale domain-specific datasets by showing that the fine-tuned models significantly outperform current SOTA VLMs in answering questions related to specific attributes, highlighting their limitations in handling complex, nuanced queries.
- 4
- Restore Anything Model via Efficient Degradation Adaptation4
A unified manner to achieve joint embedding is proposed by leveraging the inherent similarities across various degradations for efficient and comprehensive restoration by leveraging the inherent similarities across various degradations.
- Learning Generative Interactive Environments By Trained Agent Exploration4
The model GenieRedux is released - an implementation based on Genie that is reproducable, scalable and adaptable to new types of environments and a variant that uses the agent's readily available actions to factor out action prediction uncertainty during validation.
- Surface Normal Clustering for Implicit Representation of Manhattan Scenes4
This work aims to leverage the geometric prior of Manhattan scenes to improve the implicit neural radiance field representations of Manhattan scenes to self-supervise the surface normals derived explicitly in the implicit neural fields.
- Prior Based Online Lane Graph Extraction from Single Onboard Camera Image4
This work tackles online Bird's-Eye-View lane graph extraction from a single onboard camera image through a transformer based Wasserstein Autoencoder and significantly improves the performance compared to state-of-the-art methods.
- Robust and Optimal Registration of Image Sets and Structured Scenes via Sum-of-Squares Polynomials4
This paper addresses the problem of registering a known structured 3D scene, typically a 3D scan, and its metric Structure-from-Motion counterpart, and yields outstanding results in terms of robustness and optimality.
- Incremental Object Detection with Prompt-Based Methods3
This paper analyzes three different prompt-based methods under a complex domain-incremental learning setting and concludes that a strong yet practical method—combining visual prompts with replaying a small portion of previous data—achieves the best results.
- 3
- FireScope: Wildfire Risk Raster Prediction with a Chain-of-Thought Oracle3
This work proposes FireScope, a VLM-based reasoning-to-generation framework that learns from both reinforcement learning and visual supervision to predict risk rasters with complementary reasoning traces, and is the first framework to demonstrate that language-based reasoning can improve generalization in visual generation.
- A Simple and Generalist Approach for Panoptic Segmentation3
This work proposes a simple generalist framework based on a deep encoder - shallow decoder architecture with per-pixel prediction, fine-tuning a massively pretrained image model with minimal additional components to achieve state-of-the-art performance among generalist methods.
- CaLDiff: Camera Localization in NeRF via Pose Diffusion3
The proposed camera localization by pose diffusion (CaLDiff) method also integrates the pose inversion steps within the diffusion process, which offers significantly better localization, thanks to the downstream refinement-aware diffusion process.
- Unsupervised Monocular Depth Reconstruction of Non-Rigid Scenes3
An unsupervised monocular framework for dense depth estimation of dynamic scenes, which jointly reconstructs rigid and nonrigid parts without explicitly modelling the camera motion, and derives a training objective that aims to opportunistically preserve pairwise distances between reconstructed 3D points.
- Model-free Consensus Maximization for Non-Rigid Shapes3
This paper forms the model-free consensus maximization as an Integer Program in a graph using ‘rules’ on measurements and provides a method to solve it optimally using the Branch and Bound (BnB) paradigm.
- 3
- Unlocking Efficient Vehicle Dynamics Modeling via Analytic World Models2
This work designs three novel task setups in which the differentiable dynamics are combined within an end-to-end computation graph not with a policy, but a state predictor, and demonstrates how differentiable simulation enables their efficient, end-to-end learning.
- 2
- OSMDA: OpenStreetMap-based Domain Adaptation for Remote Sensing VLMs2
OSMDA is a self-contained domain adaptation framework that requires no manual labeling and no stronger external VLM teacher, and suggests that, given a strong foundation model, alignment with crowd-sourced geographic data is a practical and scalable path towards remote sensing domain adaptation.
- Partial CLIP is Enough: Chimera-Seg for Zero-shot Semantic Segmentation2
Chimera-Seg is proposed, which integrates a segmentation backbone as the body and a CLIP-based semantic head as the head, like the Chimera in Greek mythology, combining spatial precision with vision-language alignment.
- 2
- Autonomous Vehicle Controllers From End-to-End Differentiable Simulation2
This work uses a differentiable simulator to leverage and design an analytic policy gradients (APG) approach to training AV controllers on the large-scale Waymo Open Motion Dataset, and finds significant improvements in performance and robustness to noise in the dynamics, as well as overall more intuitive human-like handling.
- SPEAR-1: Scaling Beyond Robot Demonstrations via 3D Understanding2
A robotic foundation model that integrates grounded 3D perception with language-instructed embodied control and boosts the reliability of embodied control beyond what is achievable with only robotic data is introduced.
- Ternary-Type Opacity and Hybrid Odometry for RGB NeRF-SLAM2
This work introduces a ternary-type opacity (TT) model, which categorizes points on a ray intersecting a surface into three regions: before, on, and behind the surface, which enables a more accurate rendering of depth, subsequently improving the performance of image warping techniques.
- Training Dynamics Aware Neural Network Optimization with Stabilization2
This work develops a regularization scheme to encourage stable training dynamics by penalizing divergent parameter updates in neural network training, and indicates that stable network training minimizes the variance of performance across different parameter initializations, and increases robustness to the choice of learning rate.
- 2
- 2
- 2
- 2
- Local and global methods for registering 2D image sets and 3D point clouds2
The proposed registration method is globally optimal and robust to outliers, based on the theory Sum-of-Squares polynomials and a Branch-and-Bound algorithm and establishes direct relationships between 2D image measurements and 3D scene voxels.
- 1
- PAWS: Perception of Articulation in the Wild at Scale from Egocentric Videos1
PAWS is proposed, a method that directly extracts object articulations from hand-object interactions in large-scale in-the-wild egocentric videos and demonstrates that the extracted articulations benefit downstream tasks, including fine-tuning 3D articulation prediction models and enabling robot manipulation.
- Advances and Innovations in the Multi-Agent Robotic System (MARS) Challenge1
The Multi-Agent Robotic System (MARS) Challenge is proposed, held at the NeurIPS 2025 Workshop on SpaVLE, where participants explore multi-agent embodied planning using vision-language models (VLMs) to coordinate tasks and policy execution to perform robotic manipulation in dynamic environments.
- B-GRTO: Bootstrapped Group Relative Tool Optimization for Referring Segmentation1
Across three challenging referring segmentation settings, B-GRTO results in substantial improvements over plain GRPO, matching or surpassing domain-specific state-of-the-art methods, demonstrating the value of unifying reinforcement learning with differentiable auxiliary objectives for reasoning-intensive segmentation.
- Inferring Compositional 4D Scenes without Ever Seeing One1
This work proposes COM4D (Compositional 4D), a method that consistently and jointly predicts the structure and spatio-temporal configuration of 4D/3D objects using only static multi-object or dynamic single object supervision using only static multi-object or dynamic single object supervision.
- From Scan to Action: Leveraging Realistic Scans for Embodied Scene Understanding1
The efficacy of this approach is demonstrated through two downstream applications: LLM-based scene editing, enabling effective LLM understanding and adaptation of the data, and robotic simulation, achieving an 87% success rate in policy learning.
- 1
- EReLiFM: Evidential Reliability-Aware Residual Flow Meta-Learning for Open-Set Domain Generalization under Noisy Labels1
This paper introduces an unsupervised two-stage evidential loss clustering method to promote label reliability awareness, and proposes a residual flow matching mechanism that models structured domain- and category-conditioned residuals, enabling diverse and uncertainty-aware transfer paths beyond interpolation-based augmentation.
- EvenNICER-SLAM: Event-based Neural Implicit Encoding SLAM1
EvenNICER-SLAM, with an inclusion of higher-frequency event image input, significantly outperforms NICE-SLAM with reduced RGB-D input frequency and suggests the potential for event cameras to improve the robustness of dense SLAM systems against fast camera motion in real-world scenarios.
- Spatially Multi-conditional Image Generation1
A novel neural architecture operating pixel-wise, which receives the available labels as input tokens to merge them in a learned homogeneous space of labels to address the problem of heterogeneity and sparsity of the spatially multi-conditional labels.
- 1
- Improving the Behaviour of Vision Transformers with Token-Consistent Stochastic Layers1
The effectiveness of the token-consistent stochasticity is demonstrated on three different applications, namely, network calibration, adversarial robustness, and feature privacy, by boosting the performance of the respective established baselines.
- 1
- –
- –
- –
- More with Less: a Large Scale Remote Sensing VLM with a Simple Recipe–
It is shown that a generally capable vision-language model can achieve competitive or state-of-the-art performance at challenging remote sensing benchmarks, provided that it is trained at sufficient scale across diverse data and tasks, and suggests that, for remote sensing VLMs, data scale is sufficient even without architectural novelty.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Self-supervised pretraining for an iterative image size agnostic vision transformer–
This work introduces a novel sequential-to-global SSL framework based on DINO's self-distillation objective, supported by an efficient integral-image patch extraction method, which enables large-scale pretraining for image-size agnostic vision encoders.
- –
- –
- –
- –
- –
- –
- –
- ProOOD: Prototype-Guided Out-of-Distribution 3D Occupancy Prediction–
ProOOD is presented, a lightweight, plug-and-play method that couples prototype-guided refinement with training-free OOD scoring that achieves state-of-the-art performance on both in-distribution 3D occupancy prediction and OOD detection.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Neural 4D Evolution under Large Topological Changes from 2D Images–
A new architecture to discretize and encode the deformation and learn the SDF and a technique to impose the temporal consistency are introduced and a rendering scheme for color prediction based on Gaussian splatting is proposed.
- –
- Piecewise Planar Hulls for Semi-Supervised Learning of 3D Shape and Pose from 2D Images–
An end-to-end training framework for intermediate 2D keypoints extraction and 3D shape and pose estimation and achieves comparable results with fully supervised state-of-the-art methods by using only half of the annotations.
- –
- –
- –
- –
- –
- –
- –
- –
- –
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.