Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks285 from public data
- Deep High-Resolution Representation Learning for Visual Recognition5,145
The superiority of the proposed HRNet in a wide range of applications, including human pose estimation, semantic segmentation, and object detection, is shown, suggesting that the HRNet is a stronger backbone for computer vision problems.
- Efficient Test-Time Model Adaptation without Forgetting740
An active sample selection criterion is proposed to identify reliable and non-redundant samples, on which the model is updated to minimize the entropy loss for test-time adaptation, and a Fisher regularizer is introduced to constrain important model parameters from drastic changes.
- Discrimination-aware channel pruning for deep neural networks642
This work investigates a simple-yet-effective method, called discrimination-aware channel pruning, to choose those channels that really contribute to discriminative power and proposes a greedy algorithm to conduct channel selection and parameter optimization in an iterative way.
- Towards Stable Test-Time Adaptation in Dynamic Wild World617
This paper proposes a sharpness-aware and reliable entropy minimization method, called SAR, for further stabilizing TTA from two aspects: 1) remove partial noisy samples with large gradients, 2) encourage model weights to go to a flat minimum so that the model is robust to the remaining noisy samples.
- Graph Convolutional Networks for Temporal Action Localization572
This paper builds an action proposal graph, where each proposal is represented as a node and their relations between two proposals as an edge and applies the GCNs over the graph to model the relations among different proposals and learn powerful representations for the action classification and localization.
- Domain-Symmetric Networks for Adversarial Domain Adaptation356
This paper proposes a new domain adaptation method called Domain-Symmetric Networks (SymNets), which is based on a symmetric design of source and target task classifiers, based on which an additional classifier is constructed that shares with them its layer neurons.
- Closed-Loop Matters: Dual Regression Networks for Single Image Super-Resolution348
A dual regression scheme is proposed by introducing an additional constraint on LR data to reduce the space of the possible functions and it is shown that this scheme can easily adapt SR models to real-world data, e.g., raw video frames from YouTube.
- Dense Regression Network for Video Grounding346
A novel dense regression network (DRN) is designed to regress the distances between the frame within the ground truth and the starting (ending) frame of the video segment described by the query to improve the video grounding accuracy.
- Towards Effective Low-Bitwidth Convolutional Neural Networks266
This paper tackles the problem of training a deep convolutional neural network with both low-precision weights and low-bitwidth activations by proposing a two-stage optimization strategy to progressively find good local minima and adopting a novel learning scheme to jointly train a full- Precision model alongside the low-Precision one.
- Attention Guided Network for Retinal Image Segmentation226
An Attention Guided Network (AG-Net) is proposed to preserve the structural information and guide the expanding operation and an attention block is introduced to exclude the noise and reduce the negative influence of background further.
- 215
- Source-free Domain Adaptation via Avatar Prototype Generation and Adaptation215
A new robust contrastive prototype adaptation strategy to align each pseudo-labeled target data to the corresponding source prototypes via contrastive learning is developed.
- 215
- Collaborative Unsupervised Domain Adaptation for Medical Image Diagnosis195
This work innovatively proposes a Collaborative Unsupervised Domain Adaptation algorithm, which conducts transferability-aware adaptation and conquers label noise in a collaborative way and theoretically analyze the generalization performance of the proposed method, and also empirically evaluate it on both medical and general images.
- Location-Aware Graph Convolutional Networks for Video Question Answering192
This work proposes to represent the contents in the video as a location-aware graph by incorporating the location information of an object into the graph construction, and proposes to use graph convolution to infer both the category and temporal locations of an action.
- Learning Sparse SVM for Feature Selection on Very High Dimensional Datasets186
Comprehensive experimental results show that the proposed method can obtain better or competitive performance compared with existing SVM-based feature selection methods in term of sparsity and generalization performance, and can effectively handle large-scale and extremely high dimensional problems.
- Discrimination-aware Network Pruning for Deep Model Compression178
This paper proposes a simple-yet-effective method called discrimination-aware channel pruning (DCP) to choose the channels that actually contribute to the discriminative power of the network, and proposes a greedy algorithm to solve the resultant problem.
- 178
- Instance Segmentation in 3D Scenes using Semantic Superpoint Tree Networks177
An end-to-end solution of Semantic Superpoint Tree Network (SSTNet) for proposing object instances from scene points, and design in SSTNet a refinement module, termed CliqueNet, to prune superpoints that may be wrongly grouped into instance proposals.
- Structured Binary Neural Networks for Accurate Image Classification and Semantic Segmentation166
This paper proposes a ``network decomposition'' strategy, named Group-Net, in which each full-precision group can be effectively reconstructed by aggregating a set of homogeneous binary branches, and shows strong generalization to other tasks.
- Generative Low-Bitwidth Data Free Quantization161
This paper proposes a knowledge matching generator to produce meaningful fake data by exploiting classification boundary knowledge and distribution information in the pre-trained model with much higher accuracy on 4-bit quantization than the existing data free quantization method.
- Heterogeneous Domain Adaptation for Multiple Classes161
An efficient multi-class heterogeneous domain adaptation method, where data from source and target domains are represented by heterogeneous features of different dimensions, to reconstruct a sparse feature transformation matrix to map the weight vector of classifiers learned from the source domain to the target domain.
- Towards ultrahigh dimensional feature selection for big data161
A new adaptive feature scaling scheme for ultrahigh-dimensional feature selection on Big Data is presented, and then reformulate it as a convex semi-infinite programming (SIP) problem to address the SIP, and an efficient feature generating paradigm is proposed.
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition145
A new real-time action recognition architecture, called Temporal Convolutional 3D Network (T-C3D), which learns video action representations in a hierarchical multi-granularity manner, demonstrating comparable recognition performance to the state-of-the-art methods.
- Weakly-Supervised Multi-Granularity Map Learning for Vision-and-Language Navigation144
A multi-granularity map, which contains both object fine-grained details and semantic classes, to represent objects more comprehensively, and a weakly-supervised auxiliary task, which requires the agent to localize instruction-relevant objects on the map.
- RSPNet: Relative Speed Perception for Unsupervised Video Representation Learning143
It is observed that the relative playback speed is more consistent with motion patterns and thus provides more effective and stable supervision for representation learning, and a new way to perceive the playback speed and exploit the relative speed between two video clips as labels is proposed.
- EPMF: Efficient Perception-Aware Multi-Sensor Fusion for 3D Semantic Segmentation139
This work investigates a collaborative fusion scheme called perception-aware multi-sensor fusion (PMF) to effectively exploit perceptual information from two modalities, namely, appearance information from RGB images and spatio-depth information from point clouds, and proposes an improved version of PMF, i.e., EPMF, which is more efficient and effective.
- 135
- Cross-Modal Relation-Aware Networks for Audio-Visual Event Localization132
A relation-aware network to leverage both audio and visual information for accurate event localization and to reduce the interference brought by the background, an audio-guided spatial-channel attention module is proposed to guide the model to focus on event-relevant visual regions.
- Cost-Sensitive Portfolio Selection via Deep Reinforcement Learning132
A novel two-stream portfolio policy network is devised to extract both price series patterns and asset correlations, while a new cost-sensitive reward function is developed to maximize the accumulated return and constrain both costs via reinforcement learning.
- Semi-Supervised Optimal Transport for Heterogeneous Domain Adaptation130
A novel semi-supervised algorithm is proposed by exploiting the theory of optimal transport (OT), a powerful tool originally designed for aligning two different distributions, to preserve the semantic consistency between heterogeneous domains by incorporating label information into the entropic Gromov-Wasserstein discrepancy.
- Generating Visually Aligned Sound From Videos120
This work introduces an innovative audio forwarding regularizer that directly considers the real sound as input and outputs bottlenecked sound features and can control the irrelevant sound component and thus prevent the model from learning an incorrect mapping between video frames and sound emitted by the object that is out of the screen.
- 118
- 104
- A Self-Supervised Gait Encoding Approach With Locality-Awareness for 3D Skeleton Based Person Re-Identification101
A self-supervised gait encoding approach that can leverage unlabeled skeleton data to learn gait representations for person re-identification via gait features within 3D skeleton sequences and a locality-aware attention mechanism and contrastive learning scheme, which aim to preserve locality-awareness on intra-sequence level and inter- sequence level respectively during self- supervised learning.
- 101
- Open-Nav: Exploring Zero-Shot Vision-and-Language Navigation in Continuous Environment with Open-Source LLMs99
Open-Nav is introduced, a novel study that explores open-source LLMs for zero-shot VLN in the continuous environment and employs a spatial-temporal chain-of-thought (CoT) reasoning approach to break down tasks into instruction comprehension, progress estimation, and decision-making.
- Graph Convolutional Module for Temporal Action Localization in Videos93
It is claimed that the relations between action units play an important role in action localization, and a more powerful action detector should not only capture the local content of each action unit but also allow a wider field of view on the context related to it to enhance action localization.
- Contrastive Neural Architecture Search with Neural Architecture Comparators91
This paper designs and learns a Neural Architecture Comparator (NAC) to compute the probability of candidate architectures being better than a baseline one and theoretically shows that learning NAC is equivalent to optimizing the ranking over architectures.
- Relation-Aware Transformer for Portfolio Policy Learning91
Under a deep reinforcement learning paradigm for portfolio selection, a novel Relation-aware Transformer (RAT) is proposed that is structurally innovated to capture both sequential patterns and asset correlations for portfolio selection.
- 87
- AlphaFin: Benchmarking Financial Analysis with Retrieval-Augmented Stock-Chain Framework83
This work releases AlphaFin datasets, combining traditional research datasets, real-time financial data, and handwritten chain-of-thought (CoT) data, which has positive impact on training LLM for completing financial analysis, and benchmarks a state-of-the-art method, called Stock-Chain, for effectively tackling the financial analysis task.
- Weighted Block-Sparse Low Rank Representation for Face Clustering in Videos83
A new method called weighted block-sparse low rank representation (WBSLRR) is proposed which considers the available prior knowledge while learning a low rank data representation, and also develops a simple but effective approach to obtain the clustering result of faces.
- 81
- $A^2$Nav: Action-Aware Zero-Shot Robot Navigation by Exploiting Vision-and-Language Ability of Foundation Models80
This paper proposes an action-aware zero-shot VLN method ($A^2$Nav) by exploiting the vision-and-language ability of foundation models and uses the advanced reasoning ability of large language models to decompose complex navigation instructions into a sequence of action-specific object navigation sub-tasks.
- Auto-Embedding Generative Adversarial Networks For High Resolution Image Synthesis79
A novel GAN is developed called auto-embedding generative adversarial network, which simultaneously encodes the global structure features and captures the fine-grained details of real images, and is able to produce high-resolution images of promising quality directly from the input noise.
- Training Quantized Neural Networks With a Full-Precision Auxiliary Module76
The proposed method achieves near lossless performance to the full-precision model by using a 4-bit detector, which is of great practical value, and evaluates the proposed method on image classification and object detection over various quantization approaches and show consistent performance increase.
- Towards Accurate Text-based Image Captioning with Content Diversity Exploration74
This work proposes a novel Anchor-Captioner method, which not only achieves SOTA performance but also generates diverse captions to describe images.
- Adaptive cost-sensitive online classification73
A family of cost-sensitive online classification algorithms with adaptive regularization with sketching technique is proposed, which significantly accelerates the computational speed with quite slight performance loss and applies to several online anomaly detection tasks from real world.
- 72
- Length-Controllable Image Captioning70
This paper devise a non-autoregressive image captioning approach that outperforms the autoregressive baselines in terms of controllability and diversity, and also significantly improves the decoding efficiency for long captions.
- 70
- From Whole Slide Imaging to Microscopy: Deep Microscopy Adaptation Network for Histopathology Cancer Image Classification70
A novel Deep Microscopy Adaptation Network (DMAN) is proposed, which can classify MSIs effectively even without MSI annotations, by reducing domain discrepancies via adversarial learning and entropy minimization, and alleviating class imbalance with sample reweighting.
- Learning Vision-and-Language Navigation from YouTube Videos68
This paper proposes to learn an agent from these videos by creating a large-scale dataset which comprises reasonable path-instruction pairs from house tour videos and pre-training the agent on it, and proposes an action-aware generator for generating instructions from unlabeled trajectories.
- Cross-Ray Neural Radiance Fields for Novel-view Synthesis from Unconstrained Image Collections68
This paper proposes Cross-Ray NeRF (CR-NeRF) that leverages interactive information across multiple rays to synthesize occlusion-free novel views with the same appearances as the images, and theoretically finds that leveraging correlation across multiple rays promotes capturing more global information.
- Riemannian Pursuit for Big Matrix Recovery67
An efficient method, called Riemannian Pursuit, that aims to address low rank matrix recovery and fixed-rank optimization problems simultaneously and substantially outperforms existing methods when applied to large-scale and ill-conditioned matrices.
- AGE challenge: Angle Closure Glaucoma Evaluation in Anterior Segment Optical Coherence Tomography65
The top-performing approach had an average Euclidean Distance of 10 pixels (10 µm) in scleral spur localization, while in the task of angle closure classification, all the algorithms achieved satisfactory performances, with two best obtaining an accuracy rate of 100%.
- Cascade Reasoning Network for Text-based Visual Question Answering64
A novel Cascade Reasoning Network (CRN) is proposed that consists of a progressive attention module (PAM) and a multimodal reasoning graph (MRG) module that aims to explicitly model the connections and interactions between texts and visual concepts.
- 63
- Learning Discriminative Correlation Subspace for Heterogeneous Domain Adaptation61
A novel HDA method to find the optimal discriminative correlation subspace for the source and target data and applies an alternating direction method of multiplier (ADMM) algorithm to address the resulting non-convex optimization problem.
- Dynamic Extension Nets for Few-shot Semantic Segmentation60
A Dynamic Extension Network (DENet) is proposed in which a classifier for the novel class is constructed by leveraging the knowledge from the base classes and the information from novel data, and a Guided Attention Module (GAM) is designed, which can be plugged into any framework to help learn class-relevant features.
- Tuning SVM parameters by using a hybrid CLPSO–BFGS algorithm59
This paper presents a hybrid strategy to combine a comprehensive learning particle swarm optimizer (CLPSO) with Broyden-Fletcher-Goldfarb-Shanno (BFGS) method for effectively tuning the SVM parameters based on the generalization bounds and can greatly improve the stability of the parameter settings.
- ASCNet: Self-supervised Video Representation Learning with Appearance-Speed Consistency56
It is observed that the consistency between positive samples is the key to learn robust video representations and two tasks to learn appearance and speed consistency are proposed, separately.
- 56
- Masked Motion Encoding for Self-Supervised Video Representation Learning55
Motivated by the fact that human is able to recognize an action by tracking objects' position changes and shape changes, this work proposes to reconstruct a motion trajectory that represents these two kinds of change in the masked regions.
- Effective Training of Convolutional Neural Networks With Low-Bitwidth Weights and Activations55
This paper tackles the problem of training a deep convolutional neural network of both low-bitwidth weights and activations and proposes three practical approaches, including progressive quantization; stochastic precision; and joint knowledge distillation to improve the network training.
- Multi-class heterogeneous domain adaptation55
An efficient method named Sparse Heterogeneous Feature Representation (SHFR) is presented for multi-class HDA to learn a sparse feature transformation between domains with multiple classes and it is shown that the estimation error of the learned transformation decreases with the increasing number of binary classifiers.
- Test-Time Learning for Large Language Models54
A Test-Time Learning (TTL) paradigm for LLMs is proposed, namely TLM, which dynamically adapts LLMs to target domains using only unlabeled test data during testing, and introduces a Sample Efficient Learning Strategy that actively selects and emphasizes high-perplexity samples for test-time updates.
- 54
- Intelligent Home 3D: Automatic 3D-House Design From Linguistic Descriptions Only54
A House Plan Generative Model (HPGM) is proposed that first translates the language input to a structural graph representation and then predicts the layout of rooms with a Graph Conditioned Layout Prediction Network and generates the interior texture with a Language Conditioned Texture GAN.
- Oversampling for Imbalanced Data via Optimal Transport54
This paper proposes an oversampling method by exploiting global geometric information of data to make synthetic samples follow a similar distribution to that of minority class samples and introduces a novel regularization based on synthetic samples that shifts the distribution of minorityclass samples according to loss information.
- LSceneLLM: Enhancing Large 3D Scene Understanding Using Adaptive Visual Preferences52
LSceneLLM is an adaptive framework that automatically identifies task-relevant areas by leveraging LLM’s visual preference for different tasks, followed by a plug-and-play scene magnifier module to capture fine-grained details in focused areas.
- 52
- Online Adaptive Asymmetric Active Learning for Budgeted Imbalanced Data52
A novel Online Adaptive Asymmetric Active (OA3) learning algorithm is proposed, based on a new asymmetric strategy (merging both the asymmetric losses and queries strategies), and second-order optimization, which is theoretically analyzed and empirically evaluated on four real-world online anomaly detection tasks.
- Recurrent Affine Transformation for Text-to-Image Synthesis51
A recurrent affine transformation (RAT) that connects all the CAT blocks with a recurrent neural network for modeling the long-term dependency between CAT blocks is proposed, which demonstrates the effectiveness of RAT but also turns out to be a useful perspective to analyze how GANs fuse conditional information.
- SkeletonNet: A Topology-Preserving Solution for Learning Mesh Reconstruction of Object Surfaces From RGB Images51
Two models are proposed, the Skeleton-Based Graph Convolutional Neural Network (SkeGCNN and SkeDISN), which respectively build upon and improve over the existing frameworks of explicit mesh deformation and implicit field learning for the downstream surface reconstruction task.
- 51
- A Real-Time Action Representation With Temporal Encoding and Deep Compression50
A new real-time convolutional architecture for action representation, called Temporal Convolutional 3D Network (T-C3D), to capture complementary information on the appearance of a single frame and the motion between consecutive frames and a new temporal encoding method to explore the temporal dynamics of the whole video.
- 50
- 50
- 48
- Digital Gonioscopy Based on Three-dimensional Anterior-Segment OCT46
The 3D DGS is effective in detecting the eyes suspected to have PACG and has the potential to be used widely in the primary eye care community for screening of the subjects at high risk of developing PACG.
- 46
- Detecting Adversarial Data by Probing Multiple Perturbations Using Expected Perturbation Score45
This paper proposes a new statistic called expected perturbation score (EPS), which is essentially the expected score of a sample after various perturbations, and theoretically proves that EPS is a proper statistic to compute the discrepancy between two samples under mild conditions.
- Towards Accurate and Compact Architectures via Neural Architecture Transformer45
A Neural Architecture Transformer (NAT) method is proposed which casts the optimization problem into a Markov Decision Process (MDP) and seeks to replace the redundant operations with more efficient operations, such as skip or null connection.
- V2C: Visual Voice Cloning43
This work designs a set of evaluation metrics, named MCD-DTW-SL, which help eval-uate the similarity between ground-truth speeches and the synthesised ones, and proposes a new task, V2C, which seeks to convert a para-graph of text to a speech with both desired voice speci-fied by a reference audio and desired emotion specified by areference video.
- FGPrompt: Fine-grained Goal Prompting for Image-goal Navigation42
This paper designs a Fine-grained Goal Prompting (FGPrompt) method for image-goal navigation that leverages fine- grained and high-resolution feature maps in the goal image as prompts to perform conditioned embedding, which preserves detailed information in thegoal image and guides the observation encoder to pay attention to goal-relevant regions.
- 42
- Multi-modality Sensor Data Classification with Selective Attention42
This paper turns this classification task into a game and applies a deep reinforcement learning scheme to dynami-cally deal with complex situations, and intro-duce a selective attention mechanism into the rein-forcement learning Scheme to focus on the crucial dimensions of the data.
- Learning Active Camera for Multi-Object Navigation40
This paper casts moving camera to a Markov Decision Process and reformulate the active camera problem as a reinforcement learning problem to address two new challenges: how to learn a good camera policy in complex environments and how to coordinate it with the navigation policy.
- 38
- 38
- 37
- CPCM: Contextual Point Cloud Modeling for Weakly-supervised Point Cloud Semantic Segmentation36
A simple yet effective Contextual Point Cloud Modeling (CPCM) method that consists of two parts: a region-wise masking (Region-Mask) strategy and a contextual masked training (CMT) method that disentangles the learning of supervised segmentation and unsupervised masked context prediction for effectively learning the very limited labeled points and mass unlabeled points, respectively.
- PM-Net: Pyramid Multi-label Network for Joint Optic Disc and Cup Segmentation36
A Mask-RCNN based paradigm to localize ONH and jointly segment OD and OC in a whole fundus image and employs a multi-label head strategy to incorporate the prior for better performance is proposed.
- 34
- Hard Sample Matters a Lot in Zero-Shot Quantization34
HAST pays more attention to hard samples when synthesizing samples and makes synthetic samples hard to fit when training quantized models, achieving performance comparable to models that are quantized with real data.
- 34
- 34
- Multi-way backpropagation for training compact deep neural networks33
This paper first investigates the supervision vanishing issue in existing backpropagation (BP) methods and proposes to address it via an effective method, called Multi-way BP (MW-BP), which relies on multiple auxiliary losses added to the intermediate layers of the network.
- 33
- Online Adaptive Asymmetric Active Learning With Limited Budgets33
A novel Online Adaptive Asymmetric Active learning algorithm is proposed, based on a new asymmetric strategy (merging both asymmetric losses and queries strategies), and second-order optimization, which significantly accelerates the computational speed with quite slight performance degradation.
- Longitudinal Macular Retinal and Choroidal Microvasculature Changes in High Myopia32
The rate of CD loss in the DCP is significantly faster in highly myopic eyes than in healthy eyes and is related to baseline AL, while the CD in the outer ring reduces faster in eyes with longer baseline AL.
- Self-Supervised Gait Encoding with Locality-Aware Attention for Person Re-Identification32
This work proposes a generic gait encoding approach that can utilize unlabeled skeleton data to learn gait representations in a self-supervised manner, and proposes Attention-based Gait Encodings (AGEs), which are built using context vectors learned by locality-aware attention, as final gait representation.
- 32
- Learning defense transformations for counterattacking adversarial examples31
This paper empirically verify the existence of defense affine transformations that restore adversarial examples and learns defense transformations to counterattack the adversarialExamples by parameterizing the affine Transformations and exploiting the boundary information of DNNs.
- Disturbance-immune weight sharing for neural architecture search31
This work proposes a new disturbance-immune update strategy for model updating that constrain the training of subsequent architectures in an orthogonal space via Orthogonal gradient descent to preserve the knowledge learned by previous architectures.
- 31
- 30
- Double Forward Propagation for Memorized Batch Normalization30
A memorized batch normalization (MBN) is proposed, which considers multiple recent batches to obtain more accurate and robust statistics and greatly reduce the sensitivity of data and significantly improve the generalization performance.
- Uncertainty-Calibrated Test-Time Model Adaptation Without Forgetting29
An Efficient Anti-Forgetting Test-Time Adaptation (EATA) method which develops an active sample selection criterion to identify reliable and non-redundant samples for test-time entropy minimization and devise a min-max entropy regularization to selectively increase and decrease predicting confidence for confidence re-calibration.
- 28
- 28
- Detecting Machine-Generated Texts by Multi-Population Aware Optimization for Maximum Mean Discrepancy26
A novel novel multi-population aware optimization method for MMD called MMD-MP, which can avoid variance increases and thus improve the stability to measure the distributional discrepancy, is proposed.
- 26
- Face anti-spoofing with cross-stage relation enhancement and spoof material perception23
A multi-stage CNN-Transformer-based framework, which extracts local features through the convolutional layer and long-distance feature relationships via self-attention, and proposes a cross-attention multi-stage feature fusion, employing semantically high-stage features to query task-relevant features in low-stage features for further cross-stage feature fusion.
- Structured Binary Neural Networks for Image Recognition23
This paper proposes a “group decomposition” strategy, termed GroupNet, which divides a network into desired groups, and extends the GroupNet for accurate semantic segmentation by embedding the rich context into the binary structure.
- 22
- Towards Robust and Efficient Cloud-Edge Elastic Model Adaptation via Selective Entropy Distillation21
A Cloud-Edge Elastic Model Adaptation (CEMA) paradigm in which the edge models only need to perform forward propagation and the edge models can be adapted online and to reduce the communication burden is established.
- Towards Lightweight Super-Resolution With Dual Regression Learning21
A dual regression compression method is proposed to reduce model redundancy in both layer-level and channel-level based on channel pruning to alleviate the first challenge and show the effectiveness of the method in obtaining accurate and efficient SR models.
- Improving Generative Adversarial Networks With Local Coordinate Coding20
This paper proposes a GAN model with local coordinate coding (LCC), termed LCCGAN, and proposes an LCC sampling method in L CCGAN to sample meaningful points from the latent manifold, and derives the generalization bound for both LccGAN and LCCorgan++ and proves that a low-dimensional input is sufficient to achieve good generalization performance.
- Facial Image-to-Video Translation by a Hidden Affine Transformation20
This work derives an inverse formulation to the affine transformation, enabling automatic inference of the facial expression intensities from videos, and could improve users' communication experience by enabling them to conveniently and creatively produce expression GIFs, which are popular art forms in online messaging and social networks.
- 20
- NMFE-SSCC: Non-negative matrix factorization ensemble for semi-supervised collective classification20
A novel non-negative matrix factorization (NMF) based SSCC algorithm, called NMF-SSCC, to effectively learn a data representation by exploiting both labeled and unlabeled data on the network, to obtain a compact representation of network data.
- Toward Long Video Understanding via Fine-Detailed Video Story Generation19
This work proposes a Bottom-up Video Interpretation Mechanism that progressively interprets video content from clips to video and introduces a Semantic Redundancy Reduction mechanism that removes redundancy at both the visual and textual levels.
- 19
- R-GAN: Exploring Human-like Way for Reasonable Text-to-Image Synthesis via Generative Adversarial Networks19
This work proposes R-GAN, which can generate reasonable images according to the given text in a human-like way, and introduces a coarse-to-fine shape generator based on a shape knowledge base.
- Deep View Synthesis via Self-Consistent Generative Network19
A novel deep generative model, called Self-Consistent Generative Network (SCGN), which synthesizes novel views from the given input views without explicitly exploiting the geometric information, and adversarial loss is introduced to improve the photo-realism of novel views.
- 18
- HiLo: Detailed and Robust 3D Clothed Human Reconstruction with High-and Low-Frequency Information of Parametric Models18
This work empirically finds that the high-frequency (HF) and low-frequency (LF) information from a parametric model has the potential to enhance geom-etry details and improve robustness to noise, respectively, and proposes HiLo, namely clothed human reconstruction with high- and low-frequency information, which contains two components.
- Cross-Device Collaborative Test-Time Adaptation18
Experiments verify that CoLA is simple but effective, which boosts the efficiency of TTA and demonstrates remarkable superiority in collaborative, lifelong, and single-domain TTA scenarios, e.g. on follower agents.
- Deep Transferring Quantization18
A learnable attentive transfer module to identify the informative channels for alignment in network quantization to obtain an accurate low-precision model is proposed and the Kullback–Leibler (KL) divergence is introduced to further help train a low-precision model.
- Localizing Optic Disc and Cup for Glaucoma Screening via Deep Object Detection Networks18
This work introduces a deep object detection network to localize OD and OC simultaneously and achieves state-of-the-art segmentation/localization results on ORIGA dataset.
- Efficient Dynamic Ensembling for Multiple LLM Experts17
This paper proposes an efficient Dynamic Ensemble Reasoning paradigm, called DER to integrate the strengths of multiple LLM experts conditioned on dynamic inputs, and develops a Knowledge Transfer Prompt that enables the subsequent LLM candidates to transfer complementary knowledge effectively.
- 17
- 16
- Frequency-Aware Autoregressive Modeling for Efficient High-Resolution Image Synthesis16
SparseVAR is introduced, a plug-and-play acceleration framework for next-scale prediction that dynamically excludes low-frequency tokens during inference without requiring additional training and achieves notable acceleration in both HART and Infinity.
- Macular and submacular choroidal microvasculature in patients with primary open-angle glaucoma and high myopia16
Assessment of the CD in the outer ring of the SCP may facilitate the diagnosis of glaucoma in eyes with HM and reduce macular CD in different layers of the retinal capillary plexus and both particularly in theouter ring ofThe 6×6 mm scans.
- 16
- Efficient Orthogonal Non-negative Matrix Factorization over Stiefel Manifold16
A method is proposed, called Nonlinear Riemannian Conjugate Gradient ONMF (NRCG-ONMF), which updates U and V alternatively and preserves the orthogonality of U while achieving fast convergence speed.
- Online Heterogeneous Transfer Learning by Weighted Offline and Online Classifiers16
This paper builds an offline classifier based on heterogeneous similarity constructed by using labeled data from a source domain and unlabeled co-occurrence data which can be easily collected from web pages and social networks and combines the offline and online classifiers by using the Hedge weighting strategy to update their weights for ensemble prediction.
- 16
- 15
- MPGL: An Efficient Matching Pursuit Method for Generalized LASSO15
This work formulate the GL problem as a convex quadratic constrained linear programming (QCLP) problem and tailor-make a cutting plane method, and proposes an efficient Matching Pursuit Generalized LASSO (MPGL) method, which overcomes issues, and is guaranteed to converge to a global optimum.
- Single-Path Bit Sharing for Automatic Loss-Aware Model Compression14
A simple yet effective method named Single-path Bit Sharing (SBS) for automatic loss-aware model compression, which considers the network pruning as a special case of quantization and provides a unified view for model pruning and quantization.
- Scalable maximum margin matrix factorization by active Riemannian subspace search14
This paper forms M3F with a known number of latent factors as the Riemannian optimization problem on a fixed-rank matrix manifold and presents a block-wise nonlinear RiemANNian conjugate gradient method to solve it efficiently.
- Daily Assistance for Amyotrophic Lateral Sclerosis Patients Based on a Wearable Multimodal Brain-Computer Interface Mouse13
A daily assistance system for ALS patients based on a wearable multimodal brain-computer interface (BCI) mouse that provides ALS patients with effective daily assistance and shows promising long-term application prospects is developed.
- 13
- Decoupled Prototype Learning for Reliable Test-Time Adaptation12
This study reveals that minimizing the classification error of each sample causes the cross-entropy loss's vulnerability to label noise to be exposed.
- 12
- Towards interpreting deep neural networks via layer behavior understanding12
An early-exit inference method is proposed to improve the performance of the multi-label classification by “monitoring” the distribution evolution for both across-layer and single-layer along the depth and training epochs of DNNs.
- 12
- Towards Accurate Facial Motion Retargeting with Identity-Consistent and Expression-Exclusive Constraints11
This work proposes an identity-consistent constraint to learn accurate identities by encouraging consistent identity prediction across multiple frames and proposes an expression-exclusive constraint to improve performance by avoiding the co-occurrence of contradictory expression units.
- Content-aware convolutional neural networks11
A Content-aware Convolution (CAC) is proposed that automatically detects the smooth windows and applies a 1 ×1 convolutional kernel to replace the original large kernel, and is able to effectively avoid the redundant computation on similar pixels.
- Towards effective deep transfer via attentive feature alignment11
An Attentive Feature Alignment method for effective domain knowledge transfer by identifying and attending on the relevant channels and spatial features between two domains by sequentially performing attentive spatial- and channel-level feature alignments between the source and target networks is proposed.
- 11
- Retinal Image Segmentation with a Structure-Texture Demixing Network11
A structure-texture demixing network (STD-Net) that can process structures and textures differently and better is designed that demonstrates the effectiveness of the proposed segmentation strategy.
- 11
- 10
- 10
- Robust 3D Semantic Occupancy Prediction With Calibration-Free Spatial Transformation9
This work proposes a calibration-free spatial transformation based on vanilla attention to implicitly model the 2D-to-3D spatial correspondence and introduces a multi-task training framework to ease the learning of spatial correspondence without explicit spatial priors and enhance the discrimination power of 2D backbones on spatial, semantic, and texture features.
- Test-Time Model Adaptation for Quantized Neural Networks9
A continual zeroth-order adaptation (ZOA) framework that enables efficient model adaptation using only two forward passes, eliminating the computational burden of existing methods and proposing a domain knowledge management scheme to store and reuse different domain knowledge with negligible memory consumption.
- Curse of High Dimensionality Issue in Transformer for Long-context Modeling9
This paper reformulate traditional probabilistic sequence modeling as asupervised learning task, enabling the separation of relevant and irrelevant tokens and providing a clearer understanding of redundancy, and proposes the Dynamic Group Attention (DGA), which leverages the group coding to explicitly reduce redundancy by aggregating less important tokens during attention computation.
- 9
- Transformer-Based Relational Inference Network for Complex Visual Relational Reasoning9
This article regards the complex referring expression comprehension (c-REF) task as the reasoning basis, in which c-REF seeks to localise a target object in an image guided by a complex query, and proposes a Transformer-based Relational Inference Network (Trans-RINet).
- 9
- 9
- Convex Matching Pursuit for Large-Scale Sparse Coding and Subset Selection9
Theoretical analysis and experimental results show that the proposed method achieves better scalability while maintaining similar or better decoding ability compared with state-of-the-art methods on large-scale problems.
- Automatic Segmentation of Cortex and Nucleus in Anterior Segment OCT Images9
This work proposes a pipeline for automatically segmenting cortex and nucleus in a 360-degree anterior segment optical coherence tomography (AS-OCT) image and designs a shape template according to the physiological structure of nucleus to refine the boundary.
- 8
- Zero-Shot Skeleton-Based Action Recognition With Prototype-Guided Feature Alignment8
This work develops an end-to-end cross-modal contrastive training framework to improve skeleton-text alignment, ensuring sufficient discrimination for skeleton features and introduces a prototype-guided text feature alignment strategy to mitigate the adverse impact of the distribution discrepancy during testing.
- Dynamic Compressing Prompts for Efficient Inference of Large Language Models8
Inspired by the progressive difficulty adjustment in curriculum learning, a Hierarchical Prompt Compression (HPC) training strategy is introduced that gradually increases the compression difficulty, enabling the DCP-Agent to learn an effective compression method that maintains information integrity.
- Open-World Drone Active Tracking with Goal-Centered Rewards8
A Curriculum-Based Training strategy that progressively enhances the tracking performance in complex environments is introduced, and a novel reinforcement learning method called GC-VAT is proposed, which aims to improve the performance of drone tracking targets in complex scenarios.
- G-NeRF: Geometry-enhanced Novel View Synthesis from Single-View Images8
A Geometry-enhanced NeRF (G-NeRF), which seeks to enhance the geometry priors by a geometry-guided multi-view synthesis approach, followed by a depth-aware training to tackle the absence of multi-view supervision for single-view images.
- 8
- Boost Test-Time Performance with Closed-Loop Inference8
A general Closed-Loop Inference (CLI) method that first devise a filtering criterion to identify those hard-classified test samples that need additional inference loops and construct looped inference, so that the original erroneous predictions on these hard test samples can be corrected with little additional effort.
- Guided M-Net for High-Resolution Biomedical Image Segmentation with Weak Boundaries8
A guide-based model is proposed, called G-MNet, which seeks to exploit edge information from guided map to guide the corresponding lower resolution outputs of biomedical image segmentation, and will be more robust to noises and blurred object boundaries.
- 7
- Calibrate the Inter-Observer Segmentation Uncertainty via Diagnosis-First Principle7
Experimental results show that the proposed Diagnosis-First segmentation Framework (DiFF) can effectively calibrate the segmentation uncertainty, and thus significantly facilitate the corresponding disease diagnosis, which outperforms previous state-of-the-art multi-rater learning methods.
- Automated Dominative Subspace Mining for Efficient Neural Architecture Search7
A novel Neural Architecture Search method via Dominative Subspace Mining (DSM-NAS) that finds promising architectures in automatically mined subspaces and reduces the search cost but also discovers better architectures than state-of-the-art methods in various benchmark search spaces.
- A protein fitness predictive framework based on feature combination and intelligent searching7
The developed scut_ProFP (Protein Fitness Predictor) is advantageous for ML in protein engineering, providing an effective approach to data‐driven protein engineering.
- Towards Efficient Task-Driven Model Reprogramming with Foundation Models7
A Task-Driven Model Reprogramming (TDMR) framework that reprograms the foundation model to project the knowledge into a proxy space, which alleviates the adverse effect of task mismatch and domain inconsistency, and reprogram the target model via progressive distillation from the proxy space to efficiently learn the knowledge from the reprogrammed foundation model.
- Contrastive Vision-Language Alignment Makes Efficient Instruction Learner7
This paper proposes CG-VLM that applies both Contrastive and Generative alignment objectives to effectively align the representation of ViT and LLM, and produces strong vision-language alignment and is an efficient instruction learner.
- Pareto-aware Neural Architecture Generation for Diverse Computational Budgets7
A Pareto-aware Neural Architecture Generator (PNAG) which only needs to be trained once and dynamically produces the Pare to optimal architecture for any given budget via inference, which not only greatly reduces the overall search cost but also improves the search results.
- 7
- 7
- 6
- 6
- 6
- Efficient Test-Time Adaptation for Super-Resolution with Second-Order Degradation and Reconstruction6
- Modular Graph Attention Network for Complex Visual Relational Reasoning6
This paper considers reasoning on complex referring expression comprehension (c-REF) task that seeks to localise the target objects in an image guided by complex queries, and proposes a novel Modular Graph Attention Network (MGA-Net), which mimics the human language understanding mechanism.
- Cartoon-to-Photo Facial Translation with Generative Adversarial Networks6
The proposed Cartoon-to-Photo facial translation with Generative Adversarial Networks (CP-GAN) for inverting cartoon faces to generate photo-realistic and related face images can generate convincing high-quality faces that satisfy both the characteristic and identity constraints of input cartoon faces.
- 6
- Learning Sparse Confidence-Weighted Classifier on Very High Dimensional Data6
This paper presents an online-batch CW learning scheme, and then presents a novel paradigm to learn sparse CW classifiers, which essentially identifies feature groups and naturally builds a block diagonal covariance structure, making it very suitable for CW learning over very high-dimensional data.
- 6
- Adapt in the Wild: Test-Time Entropy Minimization with Sharpness and Feature Regularization5
This paper proposes a sharpness-aware and reliable entropy minimization method, called SAR, for stabilizing TTA from two aspects: to remove partial noisy samples with large gradients, and encourage model weights to go to a flat minimum so that the model is robust to the remaining noisy samples.
- 5
- 5
- Comprehensive Learning Gene Expression Programming for Automatic Implicit Equation Discovery5
This paper proposes a new mechanism named Comprehensive Learning Fitness Evaluation Mechanism (CL-FEM), which learns knowledge from disturbed information collected from several previously generated stochastic datasets, to check the validity of the equation model.
- On the Flatness of Loss Surface for Two-layered ReLU Networks.5
It is proved that two-layered ReLU networks can keep the loss of a critical point invariant, thus can incur flat regions and how to escape from flat regions is vital in training neural networks.
- Instance-level Visual Active Tracking with Occlusion-Aware Planning4
OA-VAT is proposed, a unified pipeline with three complementary modules that delivers robust, real-time performance for practical deployment of visual active tracking in real-world deployment.
- Enhancing Perception Capabilities of Multimodal LLMs with Training-Free Fusion4
VisionFuse is introduced, a novel integration framework that efficiently utilizes multiple vision encoders from off-the-shelf MLLMs to enhance visual perception without requiring additional training, significantly reducing deployment overhead.
- 4
- 4
- 4
- Structure-aware Mathematical Expression Recognition with Sequence-Level Modeling4
This work proposes a Structure-Aware Sequence-Level (SASL) model for MER to better represent and recognize the visual content of formula images, and proposes a structure-aware module to capture the relationship among different symbols.
- How to Train Your Agent to Read and Write4
A Deep ReAder-Writer (DRAW) network, which consists of a Reader that can extract knowledge graphs from input paragraphs and discover potential knowledge, a graph-to-text Writer that generates a novel paragraph, and a Reviewer that reviews the generated paragraph from three different aspects.
- 4
- 4
- Towards Large-scale and Ultrahigh Dimensional Feature Selection via Feature Generation4
By reformulating the resultant AFS model to semi-infinite programming problem, a novel feature generating method is presented to identify the most relevant features for classification problems and has the advantage of solving extremely high-dimensional and large-scale problems.
- CoNav: A Benchmark for Human-Centered Collaborative Navigation3
This work designs a novel LLM-based humanoid animation generation framework and proposes an intention-aware agent for reasoning both long-term and short-term human intention, which predicts navigation action based on the predicted intention and panoramic observation.
- Imbalance-Agnostic Source-Free Domain Adaptation via Avatar Prototype Alignment3
A new Target-aware Contrastive Prototype Generation and Adaptation (T-CPGA) method is proposed, for better prototype adaptation in the imbalance-agnostic SF-UDA task, where the class distributions of both the unseen source domain and unlabeled target domain are unknown and could be arbitrarily skewed.
- EvidFuse: Writing-Time Evidence Learning for Consistent Text-Chart Data Reporting2
EvidFuse is proposed, a training-free multi-agent framework that enables writing-time text-chart interleaved generation for data-driven reports and attains the top rank in both LLM-as-a-judge and human evaluations on chart quality, chart-text alignment, and report-level usefulness.
- Action-and-object Aware Alignment for Partially Relevant Video Retrieval2
A dual-branch framework designed to enhance retrieval by improving the modeling of action-object relationships is introduced, which uses action- and object-aware contrastive losses to enhance the model's sensitivity to action-object distinctions in the text query.
- NaVLA$^2$: A Vision-Language-Audio-Action Model for Multimodal Instruction Navigation2
This work introduces MINav, a novel task where the navigation path is precisely described by a multimodal instruction, and proposes NaVLA^2, a vision-language-audio-action model that additionally integrates spatial audio and employs a CoThinkAct module to jointly generate high-level reasoning and consistent low-level actions.
- 2
- 2
- DCIR: Dynamic Consistency Intrinsic Reward for Multi-Agent Reinforcement Learning2
This paper introduces dynamic consistency intrinsic reward (DCIR) to stimulate agents to be aware of others' behaviors and determine whether to be consistent with them, and devise a dynamic scale network (DSN) that provides learnable scale factors for the agent to dynamically ascertain whether to award consistent behavior and the magnitude of rewards.
- 2
- 2
- 2
- Towards Accurate Low Bit-Width Quantization with Multiple Phase Adaptations2
As MPA does not introduce hyperparameters that depend on different models or bit-width, the framework can be conveniently applied to various models and achieves higher accuracy than most existing methods on classification tasks for AlexNet, VGG-16 and ResNet.
- 2
- 2
- 1
- SUGAR: Learning Skeleton Representation with Visual-Motion Knowledge for Action Recognition1
This paper introduces a novel paradigm named learning Skeleton representation with visual-motion knowledge for Action Recognition (SUGAR), and proposes to supervise skeleton learning through this prior knowledge to yield discrete representations.
- 1
- Auto-3D-house Design from Structured User Requirements1
This work carefully design a structured list that divides the requirements into three parts, which focus on the attributes of rooms, the outline of the building, and the style of decoration, respectively, and proposes an automatic house design framework, called auto-3D-house design (A3HD).
- 1
- 1
- 1
- Course Assessment Reform Based on Internet and Mobile Internet Technology1
Considering the large number of students and the multitude of knowledge points in the Introduction to Computer and Software Engineering course, the smart teaching tool Yuketang is used to implement process-oriented assessments, while the summative evaluations are completed using an online examination system developed by the teaching team.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- Rethinking Binary Neural Network for Accurate Image Classification and Semantic Segmentation1
This paper proposes to train a network with both binary weights and binary activations, designed specifically for mobile devices with limited computation capacity and power consumption, and claims that considering both value and structure approximation should be the future development direction of BNNs.
- 1
- 1
- Is Matching Pursuit Solving Convex Problems1
A novel convex relaxation model is presented, which is solved by a general matching pursuit (GMP) algorithm under the convex programming framework and achieves better performance than other methods in terms of sparse recovery ability and efficiency.
- –
- Multicenter fine annotated surgical video dataset for minimally invasive glaucoma surgery–
The first multicenter, large-scale, fine-annotated surgical video dataset for MIGS, which aims to facilitate the development of advanced computer-assisted interventions that can improve the precision and safety of glaucoma surgeries.
- Zero-source LLM Hallucination Detection with Human-like Criteria Probing–
Extensive experiments show that HCPD consistently outperforms state-of-the-art baselines, offering an effective and explainable solution for zero-source hallucination detection.
- Zeta: Dual Whitening for Matrix Optimization via Coordinate-Adaptive Preconditioning–
Zeta is proposed, a dual whitening optimizer that applies coordinate whitening and spectral whitening in a strictly ordered pipeline, and it is proved that this dual pipeline strictly reduces orthogonalization error relative to pure spectral methods by improving the condition number of the input.
- –
- –
- Precedent-Informed Reasoning: Mitigating Overthinking in Large Reasoning Models via Test-Time Precedent Learning–
Experiments across mathematical reasoning, scientific QA, and code generation demonstrate that PIR consistently shortens reasoning traces while maintaining or improving final accuracy across LLMs, yielding outstanding accuracy-efficiency trade-offs.
- –
- –
- –
- –
- –
- –
- FAM: Fine-Grained Alignment Matters in Multimodal Embedding Learning with Large Vision-Language Models–
A simple yet effective Fine-grained Alignment Matters (FAM) method to achieve fine-grained vision-language embedding learning with LVLMs, and proposes Multi-granularity Aligned Contrastive (MAC) to explicitly learn and align fine-grained modality representations at multiple granularity levels using image-text pairs.
- Beyond Model Scaling: Test-Time Intervention for Efficient Deep Reasoning–
Think-with-Me is a novel test-time interactive reasoning paradigm that introduces external feedback intervention into the reasoning process, finding that transitional conjunctions serve as natural points for intervention, signaling phases of self-validation or exploration and using transitional words appropriately to prolong the reasoning enhances performance, while excessive use affects performance.
- –
- Adversarial Purification by Consistency-Aware Latent Space Optimization on Data Manifolds–
This paper reveals that samples generated by a well-trained generative model are close to clean ones but far from adversarial ones, and proposes Consistency Model-based Adversarial Purification (CMAP), which optimizes vectors within the latent space of a pre-trained consistency model to generate samples for restoring clean data.
- –
- –
- –
- –
- –
- –
- CHRIS: Clothed Human Reconstruction with Side View Consistency–
This work introduces Clothed Human Reconstruction with Side View Consistency, namely CHRIS, which consists of a Side-View Normal Discriminator that enhances global visual reasonability by distinguishing the generated side-view normals from the ground truth ones and a Multi-to-One Gradient Computation that ensures local surface consistency.
- –
- –
- –
- –
- –
- MAGIC: Map-Guided Few-Shot Audio-Visual Acoustics Modeling–
This work presents a *map-guided* framework by constructing acoustic-related visual semantic feature maps of the scenes by generating a **scene semantic map** via diffusing features and anticipating the observation semantic map.
- Likelihood-Based Text-to-Image Evaluation with Patch-Level Perceptual and Semantic Credit Assignment–
This paper proposes to evaluate text-to-image generation performance by directly estimating the likelihood of the generated images using a pre-trained likelihood-based text- to-image generative model, i.e., a higher likelihood indicates better perceptual quality and better text-image alignment.
- Purely Image-based Vault Prediction with Domain Prior Supervision for Intraocular Lens Implantation–
This work designs an automatic deep learning-based method for vault prediction only considering anterior segment optical coherence tomography (AS-OCT) images and introduces domain prior supervision into the deep learning framework.
- –
- –
- –
- –
- –
- –
- –
- –
- –
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.