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
- José M. F. MouraSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works207 from public data
- Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting9,210
An efficient transformer-based model for LSTF, named Informer, with three distinctive characteristics: a ProbSparse self-attention mechanism, which achieves O(L log L) in time complexity and memory usage, and has comparable performance on sequences' dependency alignment.
- ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate1,015
A multi-agent referee team called ChatEval is constructed to autonomously discuss and evaluate the quality of generated responses from different models on open-ended questions and traditional natural language generation (NLG) tasks, offering a human-mimicking evaluation process for reliable assessments.
- Adversarial Multiple Source Domain Adaptation638
This paper proposes multisource domain adversarial networks (MDAN) that approach domain adaptation by optimizing task-adaptive generalization bounds and conducts extensive experiments showing superior adaptation performance on both classification and regression problems: sentiment analysis, digit classification, and vehicle counting.
- SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference451
A text-guided training-free token optimization mechanism dubbed SparseVLM that eliminates the need of extra parameters or fine-tuning costs, and introduces a rank-based strategy to adaptively determine the sparsification ratio for each layer alongside a token recycling method that compresses pruned tokens into more compact representations.
- Sample-Efficient Deep Learning for COVID-19 Diagnosis Based on CT Scans440
An Self-Trans approach is proposed, which synergistically integrates contrastive self-supervised learning with transfer learning to learn powerful and unbiased feature representations for reducing the risk of overfitting in COVID-19.
- Q-Diffusion: Quantizing Diffusion Models361
This work proposes a novel PTQ method specifically tailored towards the unique multi-timestep pipeline and model architecture of the diffusion models, which compresses the noise estimation network to accelerate the generation process.
- Instance Adaptive Self-training for Unsupervised Domain Adaptation356
To effectively improve the quality of pseudo-labels, a novel pseudo-label generation strategy with an instance adaptive selector is developed and the region-guided regularization to smooth the pseudo- label region and sharpen the non-pseudo-label region is proposed.
- A Review of Single-Source Deep Unsupervised Visual Domain Adaptation349
This article review the latest single-source deep unsupervised DA methods focused on visual tasks and discusses new perspectives for future research, including discrepancy-based methods, adversarial discriminative methods, and self-supervision- based methods.
- PointCLIP V2: Prompting CLIP and GPT for Powerful 3D Open-world Learning293
This paper first collaborate CLIP and GPT to be a unified 3D open-world learner, named as Point-CLIP V2, which fully unleashes their potential for zero-shot 3D classification, segmentation, and detection, demonstrating the generalization ability for unified 3D open-world learning.
- Multi-Source Distilling Domain Adaptation257
This paper proposes a novel multi-source distilling domain adaptation (MDDA) network, which not only considers the different distances among multiple sources and the target, but also investigates the different similarities of the source samples to the target ones.
- Outlier Suppression: Pushing the Limit of Low-bit Transformer Language Models241
It is discovered that $\boldsymbol \gamma$ in LayerNorm (LN) acts as a sinful amplifier for the outliers, and the importance of outliers varies greatly where some outliers provided by a few tokens cover a large area but can be clipped sharply without negative impacts.
- 229
- FCN-rLSTM: Deep Spatio-Temporal Neural Networks for Vehicle Counting in City Cameras225
A novel FCN-rLSTM network to jointly estimate vehicle density and vehicle count is designed by connecting fully convolutional neural networks (FCN) with long short term memory networks (L STM) in a residual learning fashion and enables refined feature representation and a novel end-to-end trainable mapping from pixels to vehicle count.
- RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete208
RoboBrain is an MLLM-based model that combines robotic and general multi-modal data, utilizes a multi-stage training strategy, and incorporates long videos and high-resolution images to improve its robotic manipulation capabilities.
- Prototypical Cross-domain Self-supervised Learning for Few-shot Unsupervised Domain Adaptation204
An end-to-end Prototypical Cross-domain Self-Supervised Learning (PCS) framework for Few-shot Unsupervised Domain Adaptation (FUDA) that not only performs cross-domain low-level feature alignment, but it also encodes and aligns semantic structures in the shared embedding space across domains.
- RenderOcc: Vision-Centric 3D Occupancy Prediction with 2D Rendering Supervision186
This work is the first attempt to train multi-view 3D occupancy models only using 2D labels, reducing the dependence on costly 3D occupancy annotations and introduces an Auxiliary Ray method to tackle the issue of sparse viewpoints in autonomous driving scenarios.
- RoboMamba: Efficient Vision-Language-Action Model for Robotic Reasoning and Manipulation157
This work introduces RoboMamba, an end-to-end robotic VLA model that leverages Mamba to deliver both robotic reasoning and action capabilities, while maintaining efficient fine-tuning and inference.
- Understanding Traffic Density from Large-Scale Web Camera Data145
The FCN based method jointly estimates vehicle density and vehicle count with a residual learning framework to perform end-to-end dense prediction, allowing arbitrary image resolution, and adapting to different vehicle scales and perspectives.
- Self-Supervised Pretraining Improves Self-Supervised Pretraining142
H Hierarchical PreTraining (HPT) is explored, which decreases convergence time and improves accuracy by initializing the pretraining process with an existing pretrained model, and provides a simple framework for obtaining better pretrained representations with less computational resources.
- 130
- Expanding the prediction capacity in long sequence time-series forecasting126
An efficient transformer-based model for LSTF, named Informer, with three distinctive characteristics, which demonstrate that Informer outperforms existing methods and provides a new solution to the LSTF problem.
- Learning Invariant Representations and Risks for Semi-supervised Domain Adaptation123
This paper proposes the first method that aims to simultaneously learn invariant representations and risks under the setting of semi-supervised domain adaptation (Semi-DA), and provides a finite sample bound for both classification and regression problems under Semi-DA.
- PiMAE: Point Cloud and Image Interactive Masked Autoencoders for 3D Object Detection117
This work proposes Pi-MAE, a self-supervised pre-training framework that promotes 3D and 2D interaction through three aspects, and designs a unique cross-modal reconstruction module to enhance representation learning for both modalities.
- VoroNav: Voronoi-based Zero-shot Object Navigation with Large Language Model116
VoroNav is introduced, a novel semantic exploration framework that proposes the Reduced Voronoi Graph to extract exploratory paths and planning nodes from a semantic map constructed in real time and surpasses existing benchmarks in both success rate and exploration efficiency.
- Draw-and-Understand: Leveraging Visual Prompts to Enable MLLMs to Comprehend What You Want114
The Draw-and-Understand framework is presented, exploring how to integrate visual prompting understanding capabilities into Multimodal Large Language Models (MLLMs) and can be easily and effectively applied to various MLLMs, such as SPHINX-X and LLaVA.
- Generalized Zero-Shot Text Classification for ICD Coding105
A fast and accurate neural CRF constituency parser to batchify the inside algorithm for loss computation by direct large tensor operations on GPU, and meanwhile avoid the outside algorithm for gradient computation via efficient back-propagation.
- Margin-Based Few-Shot Class-Incremental Learning with Class-Level Overfitting Mitigation101
A novel margin-based FSCIL method is proposed to mitigate the class-level overfitting (CO) problem by providing the pattern learning process with extra constraint from the margin- based patterns themselves, and achieves state-of-the-art performance.
- MSINet: Twins Contrastive Search of Multi-Scale Interaction for Object ReID99
This work proposes a novel Twins Contrastive Mechanism (TCM) to provide more appropriate supervision for ReID architecture search, reduces the category overlaps between the training and validation data, and assists NAS in simulating real-world ReID training schemes.
- Unsupervised Domain Adaptive 3D Detection with Multi-Level Consistency96
A novel and unified framework, Multi-Level Consistency Network (MLC-Net), which employs a teacher-student paradigm to generate adaptive and reliable pseudo-targets and out-performs existing state-of-the-art methods on standard benchmarks.
- NoisyQuant: Noisy Bias-Enhanced Post-Training Activation Quantization for Vision Transformers91
NoisyQuant, a quantizer-agnostic enhancement for the post-training activation quantization performance of vision transformers, makes a surprising theoretical discovery that for a given quantizer, adding a fixed Uniform noisy bias to the values being quantized can significantly reduce the quantization error under provable conditions.
- Contrastive Multimodal Fusion with TupleInfoNCE89
This paper proposes a novel contrastive learning objective, TupleInfoNCE, which contrasts tuples based not only on positive and negative correspondences, but also by composing new negative tuples using modalities describing different scenes.
- A multimodal physiological dataset for driving behaviour analysis78
This study involved the analysis of driving behavior using multimodal physiological data collected from 35 participants, which included 59-channel EEG, single-channel ECG, 4-channel EMG, single-channel EMG, and eye movement data obtained via a six-degree-of-freedom driving simulator.
- Multiple Source Domain Adaptation with Adversarial Learning74
A new generalization bound for domain adaptation when there are multiple source domains with labeled instances and one target domain with unlabeled instances is proposed, which does not require expert knowledge about the target distribution, nor the optimal combination rule for multisource domains.
- PAD: A Dataset and Benchmark for Pose-agnostic Anomaly Detection68
This work builds MAD using 20 complex-shaped LEGO toys including 4K views with various poses, and high-quality and diverse 3D anomalies in both simulated and real environments and proposes a novel method OmniposeAD, trained using MAD, specifically designed for pose-agnostic anomaly detection.
- FreeKD: Knowledge Distillation via Semantic Frequency Prompt66
The Frequency Knowledge Distillation method is renamed as FreeKD, which determines the optimal localization and extent for the frequency distillation, and employs a position-aware relational frequency loss for dense prediction tasks, delivering a high-order spatial enhancement to the student model.
- 65
- MC-LLaVA: Multi-Concept Personalized Vision-Language Model58
Comprehensive experiments demonstrate that MC-LLaVA achieves impressive multi-concept personalized responses, paving the way for VLMs to become better user assistants.
- 53
- P2FEViT: Plug-and-Play CNN Feature Embedded Hybrid Vision Transformer for Remote Sensing Image Classification52
A hybrid architecture of convolution neural network (CNN) and ViT is proposed to improve the RSIC ability, called P2FEViT, which integrates plug-and-play CNN features with ViT, and a novel approach embedding CNN features into the ViT architecture is proposed, which can make the model synchronously capture and fuse global context and local multimodal information to further improve the classification capability of ViT.
- 52
- AffordGrasp: In-Context Affordance Reasoning for Open-Vocabulary Task-Oriented Grasping in Clutter45
AffordGrasp is a novel open-vocabulary grasping framework that leverages the reasoning capabilities of vision-language models (VLMs) for in-context affordance reasoning and infers tasks directly from implicit user instructions, enabling more intuitive and seamless human-robot interaction in everyday scenarios.
- Cross-Domain Sentiment Classification with Contrastive Learning and Mutual Information Maximization43
This work proposes CLIM: Contrastive Learning with mutual Information Maximization, the first to adopt contrastive learning for natural language processing (NLP) tasks across domains, and introduces mutual information maximization apart from CL to exploit the features that best support the final prediction.
- MoVE-KD: Knowledge Distillation for VLMs with Mixture of Visual Encoders42
A novel framework that distills the unique proficiencies of multiple vision encoders into a single, efficient encoder model and employs low-rank adaptation and mixture-of-experts to selectively activate specialized knowledge based on input features, enhancing both adaptability and efficiency.
- UniOcc: Unifying Vision-Centric 3D Occupancy Prediction with Geometric and Semantic Rendering41
This work proposes a novel Unifying Occupancy (UniOcc) prediction method, explicitly imposing spatial geometry constraint and complementing fine-grained semantic supervision through volume ray rendering, and introduces a Depth-aware Teacher Student (DTS) framework to enhance prediction accuracy using unlabeled data.
- Efficient Deweahter Mixture-of-Experts with Uncertainty-Aware Feature-Wise Linear Modulation40
This work proposes an efficient MoE architecture with weight sharing across the experts, and proposes an Uncertainty-aware Router (UaR) to assign task-specific features to different FM modules with well-calibrated weights that enables MoFME to effectively learn diverse expert functions for multiple tasks.
- 38
- RoboOS: A Hierarchical Embodied Framework for Cross-Embodiment and Multi-Agent Collaboration37
RoboOS, the first open-source embodied system built on a Brain-Cerebellum hierarchical architecture, enables a paradigm shift from single-agent to multi-agent intelligence, and enhances edge-cloud communication and cloud-based distributed inference to facilitate high-frequency interactions and enable scalable deployment.
- EmbodiedOcc++: Boosting Embodied 3D Occupancy Prediction with Plane Regularization and Uncertainty Sampler34
Improved edge accuracy and retains more geometric details while ensuring computational efficiency, which is essential for online embodied perception, is demonstrated.
- 34
- TIGeR: Tool-Integrated Geometric Reasoning in Vision-Language Models for Robotics33
This work presents TIGeR (Tool-Integrated Geometric Reasoning), a novel framework that transforms VLMs from perceptual estimators to geometric computers by enabling them to generate and execute precise geometric computations through external tools.
- Compositional Few-Shot Recognition with Primitive Discovery and Enhancing33
This work proposes an approach to FSL to learn a feature representation composed of important primitives, which is jointly trained with two parts, i.e. primitive discovery and primitive enhancing, which achieves the state-of-the-art performance on all these datasets and shows better interpretability.
- 33
- 33
- Video-CoT: A Comprehensive Dataset for Spatiotemporal Understanding of Videos Based on Chain-of-Thought32
Video-CoT, a groundbreaking dataset designed to enhance spatiotemporal understanding using Chain-of-Thought(CoT) methodologies, is introduced, providing a solid foundation for evaluating spatiotemporal understanding in video comprehension.
- Cloud-Device Collaborative Learning for Multimodal Large Language Models32
A Cloud-Device Collaborative Contin-ual Adaptation framework, designed to enhance the performance of compressed, device-deployed MLLMs by lever-aging the robust capabilities of cloud-based, larger-scale MLLMs, and validate the feasibility of the approach to real-world experiments.
- 32
- From Language to Locomotion: Retargeting-free Humanoid Control via Motion Latent Guidance30
RoboGhost is presented, a retargeting-free framework that directly conditions humanoid policies on language-grounded motion latents and enables a diffusion-based policy to denoise executable actions directly from noise, preserving semantic intent and supporting fast, reactive control.
- Training-free Regional Prompting for Diffusion Transformers30
This report proposes and implements regional prompting for FLUX based on attention manipulation, which enables DiT with fined-grained compositional text-to-image generation capability in a training-free manner.
- Integer or Floating Point? New Outlooks for Low-Bit Quantization on Large Language Models30
The Mixture of Formats Quantization (MoFQ) method achieves better or comparable results over current methods in weight-only and weight-activation post-training quantization scenarios across various tasks, with no additional hardware overhead.
- Multi-Agent Reinforcement Learning30
According to the optimization problem for each agent, equilibrium concepts are put forward to regulate the distributive behaviors of multiple agents to find the optimal multi-agent reinforcement learning policy for each agent.
- FastDriveVLA: Efficient End-to-End Driving via Plug-and-Play Reconstruction-based Token Pruning29
FastDriveVLA is proposed, a novel reconstruction-based vision token pruning framework designed specifically for autonomous driving that includes a plug-and-play visual token pruner called ReconPruner, which prioritizes foreground information through MAE-style pixel reconstruction.
- Weakly-Supervised Emotion Transition Learning for Diverse 3D Co-Speech Gesture Generation29
This work proposes a novel weakly supervised training strategy to encourage authority gesture transitions and devise an emotion mixture mechanism that provides weak supervision based on a learnable mixed emotion label for transition gestures.
- 28
- EVA: An Embodied World Model for Future Video Anticipation27
The Reflection of Generation (RoG) is proposed, a set of intermediate reasoning strategies designed to enhance video prediction that leverages the complementary strengths of pre-trained vision-language and video generation models, enabling them to function as a world model in embodied scenarios.
- 27
- 27
- H2R: A Human-to-Robot Data Augmentation for Robot Pre-training from Videos26
H2R improves the generalization ability of robotic policies by mitigating the visual discrepancies between human and robot domains, and demonstrates the potential of H2R in cross-embodiment generalization and its compatibility with vision-language-action models.
- Any2Point: Empowering Any-modality Large Models for Efficient 3D Understanding26
This paper introduces Any2Point, a parameter-efficient method to empower any-modality large models (vision, language, audio) for 3D understanding and inserts an any-to-3D guided adapter module for parameter-efficient fine-tuning.
- Multimodal large language models for bioimage analysis26
An overview of multimodal large language models through the lens of bioimage analysis is given and how these models could be built as a community to facilitate biology research is discussed.
- COLE: A Hierarchical Generation Framework for Multi-Layered and Editable Graphic Design26
This C OLE system can transform a straightforward intention prompt into a high-quality graphic design, while also supporting flexible editing based on user input, and streamline the complex process and significantly enhance generation reliability.
- PINNsAgent: Automated PDE Surrogation with Large Language Models25
PINSAgent is introduced, a novel surrogation framework that leverages large language models (LLMs) and utilizes PINNs as a foundation to bridge the gap between domain-specific knowledge and deep learning and significantly improving the accuracy of PINNs-based solutions.
- 25
- TCGM: An Information-Theoretic Framework for Semi-supervised Multi-modality Learning25
By maximizing TC-induced loss (namely TC gain) over classifiers of all modalities, these classifiers can cooperatively discover the equivalent class of ground-truth classifiers; and identify the unique ones by leveraging limited percentage of labeled data.
- 24
- 24
- DesignEdit: Multi-Layered Latent Decomposition and Fusion for Unified & Accurate Image Editing23
This work adopts the concept of layers from the design domain to manipulate objects flexibly with various operations to unify various spatial-aware image editing abilities into one framework and demonstrates that this approach consistently surpasses the latest spatial editing methods, including Self-Guidance and DiffEditor.
- Triplet Attention: Rethinking the Similarity in Transformers23
The Triplet Attention (A3) for Transformer is defined, which introduces triplet connections as the complementary dependency and allows the self-attention mechanism to attend to diverse triplets and capture complex dependency.
- MAVIS: Mathematical Visual Instruction Tuning with an Automatic Data Engine22
This paper proposes MAVIS, the first MA thematical VIS ual instruction tuning paradigm for MLLMs, involving a series of mathematical visual datasets and specialized MLLMs, and introduces MAVIS-Instruct, which is adopted to instruct-tune the MLLM for robust mathematical reasoning skills.
- Stag-1: Towards Realistic 4D Driving Simulation with Video Generation Model22
A Spatial-Temporal simulAtion for drivinG (Stag-1) model to reconstruct real-world scenes and design a controllable generative network to achieve 4D simulation, which shows promising performance in multi-view scene consistency, background coherence, and accuracy.
- A Self-Correcting Vision-Language-Action Model for Fast and Slow System Manipulation21
Self-Corrected (SC)-MLLM agent significantly improve manipulation accuracy compared to previous state-of-the-art robotic MLLM (ManipLLM), increasing from 57% to 79% on seen object categories and from 47% to 69% on unseen novel categories.
- Revisiting Mid-Level Patterns for Cross-Domain Few-Shot Recognition21
A challenging subset of CDFSL where the novel classes are in distant domains from base classes is studied, by revisiting the mid-level features, which are more transferable yet under-explored in main stream FSL work, and a residual-prediction task is proposed to boost the discriminability of mid- level features.
- 19
- MMTrail: A Multimodal Trailer Video Dataset with Language and Music Descriptions18
This work presents MMTrail, a large-scale multi-modality video-language dataset incorporating more than 20M trailer clips with visual captions, and 2M high-quality clips with multimodal captions incorporating various modality annotations, and proposes a systemic captioning framework.
- Distribution-Aware Continual Test-Time Adaptation for Semantic Segmentation18
A distribution-aware tuning (DAT) method is proposed to make the semantic segmentation CTTA efficient and practical in real-world applications and introduces the Parameter Accumulation Update (PAU) strategy to collect the updated DSP and TRP in target domain sequences.
- EMD: Explicit Motion Modeling for High-Quality Street Gaussian Splatting17
Explicit Motion Decomposition (EMD) is proposed, which models the motions of dynamic objects by introducing learnable motion embeddings to the Gaussians, enhancing the decomposition in street scenes.
- Exploring Generalizable Distillation for Efficient Medical Image Segmentation17
Generalizable Knowledge Distillation (GKD) is proposed, a novel framework for enhancing the performance of lightweight networks on cross-domain medical segmentation by generalizable knowledge distillation from powerful teacher networks.
- QD-BEV : Quantization-aware View-guided Distillation for Multi-view 3D Object Detection17
The method QD-BEV enables a novel view-guided distillation (VGD) objective, which can stabilize the quantization-aware training (QAT) while enhancing the model performance by leveraging both image features and BEV features.
- 16
- CoCoGesture: Towards coherent co-speech 3D gesture generation in the wild15
This paper proposes CoCoGesture, a novel framework enabling coherent and diverse gesture synthesis from unseen human speech prompts, built upon the custom-designed pretrain-fintune training paradigm.
- 15
- 15
- 14
- METIS: Multi-Source Egocentric Training for Integrated Dexterous Vision-Language-Action Model14
This work proposes METIS, a vision-language-action (VLA) model for dexterous manipulation pretrained on multi-source egocentric datasets, which demonstrates exceptional dexterous manipulation capabilities and highlights METIS as a promising step toward a generalist model for dexterous manipulation.
- Towards Unifying Understanding and Generation in the Era of Vision Foundation Models: A Survey from the Autoregression Perspective14
This survey is the first survey to comprehensively summarize autoregressive vision foundation models under the trend of unifying understanding and generation and categorizes autoregressive vision foundation models from their vision tokenizers and autoregression backbones.
- CSQ: Growing Mixed-Precision Quantization Scheme with Bi-level Continuous Sparsification14
Continuous Sparsification Quantization (CSQ) stabilizes the bit-level mixed-precision training process with a bi-level gradual continuous sparsification on both the bit values of the quantized weights and the bit selection in determining the quantization precision of each layer.
- DualVLA: Building a Generalizable Embodied Agent via Partial Decoupling of Reasoning and Action13
This work introduces a dual-layer data pruning method that removes redundant embodied reasoning, preventing it from adversely influencing action learning, and designs a dual-teacher adaptive distillation strategy that assigns different supervision signals to different data domains while maintaining reasoning ability.
- 13
- Biphasic Face Photo-Sketch Synthesis via Semantic-Driven Generative Adversarial Network With Graph Representation Learning13
A novel semantic-driven generative adversarial network to address the above issues, cooperating with graph representation learning, and presents a biphasic interactive cycle training strategy that outperforms the state-of-the-art competitors on the CUHK Face Sketch (CUFS) and CUHK Face Sketch FERET (CUFSF) datasets.
- Can World Models Benefit VLMs for World Dynamics?12
This work strives to investigate the capabilities when world model priors are transferred into Vision-Language Models, and empirically investigates this class of models, finding that generative encoders can capture latents useful for downstream understanding that show distinctions from conventional encoders.
- 12
- 12
- Efficient Deweather Mixture-of-Experts with Uncertainty-aware Feature-wise Linear Modulation12
This work proposes an efficient MoE architecture with weight sharing across the experts, and proposes an Uncertainty-aware Router (UaR) to assign task-specific features to different FM modules with well-calibrated weights that enables MoFME to effectively learn diverse expert functions for multiple tasks.
- Annotation-Efficient Untrimmed Video Action Recognition12
This paper targets a new problem, Annotation-Efficient Video Recognition, to reduce the requirement of annotations for both large amount of samples and the action location, and proposes an open-set detection based method to find the NBG and FG, and a contrastive learning method to learn IBG and distinguish NBG in a self-supervised way.
- SliceOcc: Indoor 3D Semantic Occupancy Prediction with Vertical Slice Representation11
A new vertical slice representation that divides the scene along the vertical axis and projects spatial point features onto the nearest pair of parallel planes is presented, setting a new state-of-the-art performance among RGB camera-based models for indoor 3D semantic occupancy prediction.
- 11
- Unsupervised Spike Depth Estimation via Cross-modality Cross-domain Knowledge Transfer11
This work proposes a cross-modality cross-domain (BiCross) framework to realize unsupervised spike depth estimation by introducing simulated mediate source spike data, and designs a Coarse-to-Fine Knowledge Distillation (CFKD) approach to facilitate comprehensive cross-modality knowledge transfer.
- Implicit neural image field for biological microscopy image compression10
This study presents a flexible AI-based method for compressing microscopy images, achieving high compression while preserving details critical for analysis, with support for task-specific optimization and arbitrary-resolution decompression.
- Proximity QA: Unleashing the Power of Multi-Modal Large Language Models for Spatial Proximity Analysis10
This work introduces Proximity Question Answering (Proximity QA), a novel framework designed to enable MLLMs to infer the proximity relationship between objects in images, and proposes a VQA dataset called Proximity-110K, containing additional instructions that incorporate depth information and the proximity relationships of objects.
- SCBench: A Sports Commentary Benchmark for Video LLMs10
A novel task: sports video commentary generation, developed on SCBench, which conducts comprehensive evaluations on multiple Video LLMs and chain-of-thought baseline methods, and provides a fresh perspective for future research, aiming to enhance models' overall capabilities in complex visual understanding tasks.
- 10
- DesignEdit: Unify Spatial-Aware Image Editing via Training-free Inpainting with a Multi-Layered Latent Diffusion Framework9
This paper designs a key-masking self-attention scheme alongside artifact suppression to achieve background inpainting within the denoising process, leveraging the powerful generative capabilities of the Latent Diffusion Model, Stable Diffusion XL-1.0.
- 9
- BEVUDA++: Geometric-Aware Unsupervised Domain Adaptation for Multi-View 3D Object Detection9
An innovative geometric-aware teacher-student framework, BEVUDA++, to diminish the issue of domain shift, comprising a Reliable Depth Teacher (RDT) and a Geometric Consistent Student (GCS) model, is introduced.
- 9
- WM-MoE: Weather-aware Multi-scale Mixture-of-Experts for Blind Adverse Weather Removal9
This work proposes a novel Mixture of Weather Experts(MoWE) Transformer framework to handle complex weather removal in a perception-aware fashion, and designs a Weather-aware Router to make the experts targeted more relevant to weather types while without the need for weather type labels during inference.
- 9
- Modeling relation paths for knowledge base completion via joint adversarial training9
A novel method to explore the intrinsic relationship between the single relation and multi-hop paths between paired entities and Experimental results show that this method outperforms existing path information-based approaches.
- DOZE: A Dataset for Open-Vocabulary Zero-Shot Object Navigation in Dynamic Environments8
A Dataset for Open-Vocabulary Zero-Shot Object Navigation in Dynamic Environments (DOZE) that comprises ten high-fidelity 3D scenes with over 18 k tasks, aiming to mimic complex, dynamic real-world scenarios, is proposed and enhanced by integrating capabilities for detecting collisions between the agent and moving obstacles.
- BEVUDA: Multi-geometric Space Alignments for Domain Adaptive BEV 3D Object Detection8
A Multi-space Alignment Teacher-Student (MATS) framework to ease the domain shift accumulation, which consists of a Depth-Aware Teacher (DAT) and a Geometric-space Aligned Student (GAS) model.
- PM-DETR: Domain Adaptive Prompt Memory for Object Detection with Transformers8
This study proposes a hierarchical Prompt Domain Memory (PDM), which comprehensively leverages the prompt memory to extract domain-specific knowledge and explicitly constructs a long-term memory space for the data distribution, which represents better domain diversity compared to existing methods.
- BEV-LGKD: A Unified LiDAR-Guided Knowledge Distillation Framework for BEV 3D Object Detection8
A unified framework named BEV-LGKD to transfer the knowledge in the teacher-student manner is proposed and the quality of depth estimation is crucial for BEV perception, and depth distillation is introduced to the framework.
- SpikeNVS: Enhancing Novel View Synthesis from Blurry Images via Spike Camera7
This study introduces a new method that uses the spike camera to overcome limitations of event-RGB approaches, and designs the Texture from Spike (TfS) loss, which handles foreground objects with backgrounds simultaneously.
- RepCaM: Re-parameterization Content-aware Modulation for Neural Video Delivery7
This paper introduces a novel Re-parameterization Content-aware Modulation (RepCaM) method to modulate all the video chunks with an end-to-end training strategy, achieving state-of-the-art results in video restoration quality and delivery bandwidth compression.
- Revisiting Mid-Level Patterns for Distant-Domain Few-Shot Recognition.7
This paper revisits distant-domain FSL, a challenging subset of CDFSL, by revisiting the mid-level features, which are more transferable yet under-explored in main stream FSL work, and proposes a residual-prediction task to boost the discriminability of mid- level features.
- 7
- 7
- Multi‐Annual Inventorying of Retrogressive Thaw Slumps Using Domain Adaptation6
This work is the first to apply an unsupervised domain adaptation to automatically map the RTSs on a multi‐annual timescale, demonstrating a strong potential of its applicability for monitoring large‐scale, multi‐temporal evolution of geomorphological features.
- High-Quality 3D Creation From a Single Image Using Subject-Specific Knowledge Prior6
This paper presents a novel two-stage approach for generating high-quality 3D models from a single image, and introduces a shading modeaware prior into the NeRF optimization process, enhancing the geometry and refining texture in the coarse outputs to achieve superior quality.
- EmpathyAgent: Can Embodied Agents Conduct Empathetic Actions?6
This work introduces EmpathyAgent, the first benchmark to evaluate and enhance agents' empathetic actions across diverse scenarios, and proposes an empathy-specific evaluation suite that evaluates the agents' empathy process.
- Learning from Mistakes: Iterative Prompt Relabeling for Text-to-Image Diffusion Model Training6
This work proposes Iterative Prompt Relabeling (IPR), a novel algorithm that aligns images to text through iterative image sampling and prompt relabeling with feedback, demonstrating superior performance compared to previous RL methods.
- When Visible Light (Backscatter) Communication Meets Neuromorphic Cameras in V2X6
This paper proposes NeuromorphicVLC, a visible light communication system equipped with neuromorphic cameras as optical receivers to improve its performance, and develops a complete signal processing pipeline to detect the VLC transmitters and demodulate the messages.
- 6
- 6
- 6
- 5
- Expert-level vision-language foundation model for real-world radiology and comprehensive evaluation5
The development of RadFound represents an advancement in radiology generalists, demonstrating broad applicability potential for integration into clinical workflows.
- MR-MLLM: Mutual Reinforcement of Multimodal Comprehension and Vision Perception5
This work proposes the Mutually Reinforced Multimodal Large Language Model (MR-MLLM), a novel framework that synergistically enhances visual perception and multimodal comprehension and vision perception synergistically and proposes the perception-enhanced cross-modal integration method.
- 5
- OmniArch: Building Foundation Model For Scientific Computing5
Om-niArch’s pre-training involves a versatile pipeline that processes multi-physics spatio-temporal data, casting forward problem learning into scalable auto-regressive tasks, while the novel Physics-Informed Reinforcement Learning (PIRL) technique during fine-tuning ensures alignment with physical laws.
- VeCAF: Vision-language Collaborative Active Finetuning with Training Objective Awareness5
VeCAF optimizes a parametric data selection model by incorporating the training objective of the model being tuned, and guides the PVM towards the performance goal with improved data and computational efficiency.
- MoEC: Mixture of Experts Implicit Neural Compression5
This work proposes MoEC, a novel implicit neural compression method based on the theory of mixture of experts that can adaptively find the optimal partition in an end-to-end manner and is able to uphold the PSNR of 48.16.
- Split-Ensemble: Efficient OOD-aware Ensemble via Task and Model Splitting5
This work proposes a novel subtask-splitting ensemble training objective, where a common multiclass classification task is split into several complementary subtasks, and builds a tree-like Split-Ensemble architecture by performing iterative splitting and pruning from a shared backbone model, where each branch serves as a submodel corresponding to a subtask.
- 5
- OmniIndoor3D: Comprehensive Indoor 3D Reconstruction4
OmniIndoor3D provides comprehensive 3D indoor scene understanding, which facilitates accurate and robust robotic navigation and achieves state-of-the-art results in appearance, geometry, and panoptic reconstruction.
- 4
- MathSticks: A Benchmark for Visual Symbolic Compositional Reasoning with Matchstick Puzzles4
A benchmark for Visual Symbolic Compositional Reasoning (VSCR), which unifies visual perception, symbolic manipulation, and arithmetic consistency, is introduced, which establishes \textsc{MathSticks} as a rigorous testbed for advancing compositional reasoning across vision and symbols.
- 4
- Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection4
This paper introduces Point-DETR3D, a teacher-student framework for weakly semi-supervised 3D detection, designed to fully capitalize on point-wise supervision within a constrained instance-wise annotation budget, and proposes an explicit positional query initialization strategy to enhance the positional prior.
- 4
- 4
- Decoupling Global and Local Representations from/for Image Generation4
This work demonstrates that with only architectural inductive biases, a generative model with a plain log-likelihood objective is capable of learning decoupled representations, requiring no explicit supervision.
- 4
- 4
- 4
- PIGEON: VLM-Driven Object Navigation via Points of Interest Selection3
PIGEON (Point of Interest Guided Exploration for Object Navigation), a VLM-driven framework that formulates object navigation as raw-observation-grounded sparse decision problem, and introduces Points of Interest (PoIs) as sparse visual decision units that couple geometrically executable waypoints with raw egocentric observations.
- 3
- ChainV: Atomic Visual Hints Make Multimodal Reasoning Shorter and Better3
ChainV is proposed, a framework that dynamically integrates visual hints into the reasoning process, thereby making multimodal reasoning shorter and better and introducing a consistency-based evaluation mechanism to assess the reliability of the chosen hint.
- PiGW: A Plug-in Generative Watermarking Framework3
Plug-in Generative Watermarking (PiGW) is proposed as a general framework for integrating watermarks into generative images and embeds watermark information into the initial noise using a learnable watermark embedding network and an adaptive frequency spectrum mask.
- RustNeRF: Robust Neural Radiance Field with Low-Quality Images3
To improve NeRF's robustness under real-world inputs, this work trains a 3D-aware preprocessing network that incorporates real-world degradation modeling and proposes a novel implicit multi-view guidance to address information loss during image degradation and restoration.
- 3
- 3
- Uncertainty Guided Depth Fusion for Spike Camera3
A novel Uncertainty-Guided Depth Fusion (UGDF) framework to fuse the predictions of monocular and stereo depth estimation networks for spike camera to achieve state-of-the-art results on CitySpike20K, surpassing all monocular or stereo spike depth estimation baselines.
- Feature Fusion for Image Retrieval With Adaptive Bitrate Allocation and Hard Negative Mining3
This paper proposes a highly accurate CDVS-guided hard negative mining scheme when training CNN embeddings and proposes a more compact CDVS descriptor database building scheme through bitrate allocation, which can reduce information redundancy and boost image retrieval performance.
- Optimising Time Series Forecasting Transformers to Linear Complexity2
This paper proposes to chunk the tokens before feeding them to the encoder, in which attention calculations between tokens are limited to within the same chunk, and reduces the complexity of the transformer‐based model from OL2 $O\left({L}^{2}\right)$ to O(L) $O(L)$ .
- 2
- Fisher-aware Quantization for DETR Detectors with Critical-category Objectives2
This work finds that certain critical categories have a higher sensitivity to quantization, and are prone to overfitting after quantization-aware training (QAT), and proposes methods to improve performance for the critical categories.
- 2
- Learning from Different Samples: A Source-free Framework for Semi-supervised Domain Adaptation2
A novel source-free framework to achieve semi-supervised fine-tuning of the source pre-trained model on the target domain, and decouples SSDA from the perspectives of different target samples, specifically designing robust learning techniques for unlabeled, reliably labeled, and noisy pseudo-labeled target samples.
- 2
- 2
- 2
- Integrating Learning and Planning2
This chapter introduces the concepts of model and model-based methods, and presents the integration architecture combining learning and planning, with detailed illustration on Dyna-Q algorithm.
- 1
- GaussianEnhancer++: A General GS-Agnostic Rendering Enhancer1
This work proposes GaussianEnhancer, a general GS-agnostic post-processor that employs a degradation-driven view blending method to improve the rendering quality of GS models and introduces GaussianEnhancer++, which features an enhanced GS-style degradation simulator, leading to improved enhancement quality.
- 1
- 1
- RepCaM++: Exploring Transparent Visual Prompt With Inference-Time Re-Parameterization for Neural Video Delivery1
This paper introduces RepCaM++, an innovative framework based on a novel Re-parameterization Content-aware Modulation (RepCaM) module that uniformly modulates video chunks that achieves state-of-the-art results in video restoration quality and delivery bandwidth compression.
- A Vanilla Multi-Task Framework for Dense Visual Prediction Solution to 1st VCL Challenge -- Multi-Task Robustness Track1
A vanilla framework named UniNet is proposed that seamlessly combines various visual perception algorithms into a multi-task model that achieves a 49.6 overall score and provides some interesting observations in experiments which may facilitate the development of multi-task learning in dense visual prediction.
- ASGDiffusion: Parallel High-Resolution Generation with Asynchronous Structure Guidance1
A novel method named ASGDiffusion is introduced for parallel HR generation with Asynchronous Structure Guidance (ASG) using pre-trained diffusion models to solve the pattern repetition problem of HR image generation and achieves state-of-the-art HR generation.
- 1
- 1
- HUB: Guiding Learned Optimizers with Continuous Prompt Tuning1
This work proposes a hybrid-update-based (HUB) optimization strategy inspired by recent advancements in hard prompt tuning and result selection techniques used in large language and vision models that is able to retain the benefits of learned optimizers while stabilizing the training process and improving testing performance.
- BadRes: Reveal the Backdoors Through Residual Connection1
This paper proposes a simple yet strong backdoor attack method called BadRes, where the residual connections play as a turnstile to be deterministic on clean inputs while unpredictable on poisoned ones, and analyzes it with state-of-the-art defense methods and reveals the fundamental weakness lying in residual connections.
- 1
- 1
- Domain-adaptive matching bridges synthetic and in vivo neural dynamics for neural circuit connectivity inference1
DeepDAM fine-tunes deep neural networks on a combined dataset of synthetic simulations and unlabeled in vivo recordings, aligning the model’s feature representations with real neural dynamics to mitigate model mismatch, and achieves near-perfect connectivity inference performance.
- 1
- Effective Ensemble of Deep Neural Networks Predicts Neural Responses to Naturalistic Videos1
An ensemble of models that extract representations based on the input videos from 4 perspectives: image streams, motion, edges, and audio are built that achieved state-of-the-art performance on both the mini track and the full track tasks.
- 1
- 1
- Hierarchical Attention Networks for Knowledge Base Completion via Joint Adversarial Training.1
This paper proposes a novel method to capture the shared features between entity pair as the basis for inferring missing relations, and develops Hierarchical Attention Networks to automatically encode the inputs into low-dimensional vectors and exploit two partial parameter-shared components.
- 1
- –
- –
- SparseStreet: Sparse Gaussian Splatting for Real-Time Street Scene Simulation–
This work introduces a node-based learnable pruning strategy that systematically removes low-contributing Gaussian primitives while preserving visually critical regions in dynamic objects while significantly reducing the total number of Gaussian primitives.
- –
- –
- –
- –
- –
- –
- –
- GPD-1: Generative Pre-training for Driving–
A unified Generative Pre-training for Driving (GPD-1) model to accomplish all these tasks altogether without additional fine-tuning, including scene generation, traffic simulation, closed-loop simulation, map prediction, and motion planning.
- Heterogenous Memory Augmented Neural Networks–
This paper introduces a novel heterogeneous memory augmentation approach for neural networks which, by introducing learnable memory tokens with attention mechanism, can effectively boost performance without huge computational overhead.
- UnrealNAS: Can We Search Neural Architectures with Unreal Data?–
This work constructs three types of unreal datasets using randomly labeled real images; generated images and labels; and generated Gaussian noise with random labels to explore if NAS needs real data, suggesting the feasibility of performing NAS with unreal data.
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.