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
- Tieniu TanSuggested from co-authorship
- Yunhong WangSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works328 from public data
- Scale-Aware Trident Networks for Object Detection1,089
A novel Trident Network (TridentNet) aiming to generate scale-specific feature maps with a uniform representational power is proposed and a parallel multi-branch architecture in which each branch shares the same transformation parameters but with different receptive fields is constructed.
- 448
- Driving Into the Future: Multiview Visual Forecasting and Planning with World Model for Autonomous Driving426
The Drive-Wm is proposed, the first driving world model compatible with existing end-to-end planning models, that enables driving into multiple futures based on distinct driving maneuvers, and determines the optimal trajectory according to the image-based rewards.
- Bottom-Up Human Pose Estimation Via Disentangled Keypoint Regression398
It is empirically show that the proposed direct regression method outperforms keypoint detection and grouping methods and achieves superior bottom-up pose estimation results on two benchmark datasets, COCO and CrowdPose.
- The Devil Is in the Details: Window-based Attention for Image Compression369
This paper finds that combining the local-aware attention mechanism with the global-related feature learning could meet the expectation in image compression, and proposes a novel Symmetrical TransFormer (STF) framework with absolute transformer blocks in the down-sampling encoder and up-sampled decoder.
- Hierarchical Convolutional Neural Networks for EEG-Based Emotion Recognition337
Benefiting from the strong representational learning capacity in the two-dimensional space, HCNN is efficient in emotion recognition especially on Beta and Gamma waves.
- GIFT: A Real-Time and Scalable 3D Shape Search Engine298
The proposed 3D shape search engine, which combines GPU acceleration and Inverted File (Twice), is named as GIFT, which outperforms the state-of-the-art methods significantly in retrieval accuracy on various shape benchmarks and competitions.
- 285
- RangeDet: In Defense of Range View for LiDAR-based 3D Object Detection277
This paper proposes an anchor-free single-stage LiDAR-based 3D object detector – RangeDet, and designs three components to address two issues overlooked by previous works: the scale variation between nearby and far away objects and the inconsistency between the 2D range image coordinates used in feature extraction and the 3D Cartesian coordinate used in output.
- Sequence Level Semantics Aggregation for Video Object Detection271
This work argues that aggregating features in the full-sequence level will lead to more discriminative and robust features for video object detection, and devise a novel Sequence Level Semantics Aggregate (SELSA) module to achieve this goal.
- SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplines266
This work presents SuperGPQA, a comprehensive benchmark that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines and management of a large-scale annotation process, involving over 80 expert annotators and an interactive Human-LLM collaborative system, offering valuable methodological guidance for future research initiatives of comparable scope.
- RoleLLM: Benchmarking, Eliciting, and Enhancing Role-Playing Abilities of Large Language Models265
RoleLLM is introduced, a framework to benchmark, elicit, and enhance role-playing abilities in LLMs, and creates RoleBench, the first systematic and fine-grained character-level benchmark dataset forrole-playing with 168,093 samples.
- DarkRank: Accelerating Deep Metric Learning via Cross Sample Similarities Transfer238
This work introduces a new type of knowledge---cross sample similarities for model compression and acceleration that can be naturally derived from deep metric learning model and brings the "learning to rank" technique intoDeep metric learning formulation.
- CIAN: Cross-Image Affinity Net for Weakly Supervised Semantic Segmentation237
This paper proposes an end-to-end cross- image affinity module, which exploits pixel-level cross-image relationships with only image-level labels and achieves a new state-of-the-art result by only using image- level labels for weakly supervised semantic segmentation.
- Learning Integral Objects With Intra-Class Discriminator for Weakly-Supervised Semantic Segmentation219
- Sparse Instance Activation for Real-Time Instance Segmentation210
Owing to the simple yet effective designs with instance activation maps, SparseInst has extremely fast inference speed and achieves 40 FPS and 37.9 AP on the COCO benchmark, which significantly out-performs the counterparts in terms of speed and accuracy.
- Sharpness-Aware Gradient Matching for Domain Generalization185
This paper presents two conditions to ensure that the model could converge to a flat minimum with a small loss, and presents an algorithm, named Sharpness-Aware Gradient Matching (SAGM), to meet the two conditions for improving model generalization capability.
- Spectral Feature Transformation for Person Re-Identification158
Inspired by spectral clustering, a novel module termed Spectral Feature Transformation is developed to facilitate the optimization of group-wise similarities and adds no burden to the inference and can be applied to various scenarios.
- Improving Pedestrian Attribute Recognition With Weakly-Supervised Multi-Scale Attribute-Specific Localization154
A flexible Attribute Localization Module (ALM) is proposed to adaptively discover the most discriminative regions and learns the regional features for each attribute at multiple levels and a feature pyramid architecture is introduced to enhance the attribute-specific localization at low-levels with high-level semantic guidance.
- RefineMask: Towards High-Quality Instance Segmentation with Fine-Grained Features148
A new method called RefineMask is proposed for high-quality instance segmentation of objects and scenes, which incorporates fine-grained features during the instance-wise segmenting process in a multi-stage manner and succeeds in segmenting hard cases such as bent parts of objects that are oversmoothed by most previous methods.
- Group-Wise Semantic Mining for Weakly Supervised Semantic Segmentation147
This work designs a graph neural network (GNN) for group-wise semantic segmentation (WSSS), wherein input images are represented as graph nodes, and the underlying relations between a pair of images are char- acterized by an efficient co-attention mechanism.
- Hard-Aware Point-to-Set Deep Metric for Person Re-identification146
In addition to person re-ID, the proposed Hard-Aware Point-to-Set (HAP2S) loss with a soft hard-mining scheme applies to generic deep metric learning benchmarks including CUB-200-2011 and Cars196, and achieves state-of-the-art results.
- 145
- DriveVLA-W0: World Models Amplify Data Scaling Law in Autonomous Driving144
DriveVLA-W0 is proposed, a training paradigm that employs world modeling to predict future images and amplifies the data scaling law, showing that performance gains accelerate as the training dataset size increases.
- Implicit Sample Extension for Unsupervised Person Re-Identification144
An Implicit Sample Extension method to generate support samples from actual samples and their neighbouring clusters in the embedding space through a progressive linear interpolation (PLI) strategy, which reduces the “sub and mixed” clustering errors, thus improving the Re-ID performance.
- Learnable Graph Matching: Incorporating Graph Partitioning with Deep Feature Learning for Multiple Object Tracking143
To make the optimization end-to-end differentiable, the original graph matching into continuous quadratic programming and then incorporate the training of it into a deep graph network with the help of the implicit function theorem is relaxed.
- 142
- 141
- 121
- 116
- 114
- Look Closer to Segment Better: Boundary Patch Refinement for Instance Segmentation109
A conceptually simple yet effective post-processing refinement framework to improve the boundary quality based on the results of any instance segmentation model, termed BPR, which yields significant improvements over the Mask R-CNN baseline on Cityscapes benchmark, especially on the boundary-aware metrics.
- Crowd counting in public video surveillance by label distribution learning109
A label distribution learning (LDL) strategy is introduced into crowd counting, where crowd images are labelled with label distributions instead of the conventional single labels, which proves the effectiveness and robustness of the method.
- 108
- Generalizing Person Re-Identification by Camera-Aware Invariance Learning and Cross-Domain Mixup107
This work improves the vanilla neighborhood invariance approach by imposing the constraint in a camera-aware manner and proposes a novel cross-domain mixup scheme that alleviates the abrupt transfer by introducing the interpolation between the two domains as a transition state.
- Hard Patches Mining for Masked Image Modeling106
This work proposes Hard Patches Mining (HPM), a brand-new framework for MIM pre-training that introduces an auxiliary loss predictor, predicting patch-wise losses first and deciding where to mask next, and empirically finds that solely introducing the loss prediction objective leads to powerful representations.
- A Survey of On-line Signature Verification102
This paper presents a review of researches on on-line signature verification during recent years and lists some of the works that provide promising results.
- CMCGAN: A Uniform Framework for Cross-Modal Visual-Audio Mutual Generation99
This paper proposes a Cross-Modal Cycle Generative Adversarial Network (CMCGAN) to handle cross-modal visual-audio mutual generation and develops a dynamic multimodal classification network to handle the modality missing problem.
- SARPNET: Shape attention regional proposal network for liDAR-based 3D object detection95
A novel 3D object detection network called SARPNET is introduced, which deploys a new low-level feature encoder to remedy the sparsity and inhomogeneity of LiDAR point clouds with an even sample method, and embodies a shape attention mechanism that learns the statistic 3D shape priors of objects and uses them to spatially enhance semantic embeddings.
- 91
- A Comparative Study on Reasoning Patterns of OpenAI's o1 Model88
This work compares o1 with existing Test-time Compute methods by using OpenAI's GPT-4o as a backbone on general reasoning benchmarks in three domains, and shows that the o1 model has achieved the best performance on most datasets.
- Boosting Decision-Based Black-Box Adversarial Attacks with Random Sign Flip82
It is shown that just randomlyipping the signs of a small number of entries in adversarial perturbations can significantly boost the attack performance.
- 80
- 79
- Image Caption Generation with Part of Speech Guidance78
Experimental results on the most popular benchmark datasets, e.g., Flickr30k and MS COCO, consistently demonstrate that the method can significantly enhance the performance of a standard image caption generation model, and achieve the conpetitive results.
- Enhancing person re-identification by integrating gait biometric78
Gait biometric is an effective feature integrated with appearance features to enhance person re-identification and is tested on CASIA gait dataset.
- CityGaussianV2: Efficient and Geometrically Accurate Reconstruction for Large-Scale Scenes77
This paper implements a decomposed-gradient-based densification and depth regression technique to eliminate blurry artifacts and accelerate convergence of 2D Gaussian Splatting, and introduces an elongation filter that mitigates Gaussian count explosion caused by 2DGS degeneration.
- Large-Scale Object Detection in the Wild From Imbalanced Multi-Labels76
This work designs a concurrent softmax to handle the multi-label problems in object detection and proposes a soft-sampling methods with hybrid training scheduler to deal with the label imbalance.
- FSD V2: Improving Fully Sparse 3D Object Detection With Virtual Voxels73
This paper presents FSDv2, an evolution that aims to simplify the previous FSDv1 and eliminate the ad-hoc heuristics in its handcrafted instance-level representation, thus promoting better universality and introducing virtual voxels, taking over the clustering-based instance segmentation in FSDv1.
- Fairly Adaptive Negative Sampling for Recommendations72
This paper proposes a Fairly adaptive Negative sampling approach (FairNeg), which improves item group fairness via adaptively adjusting the group-level negative sampling distribution in the training process, and shows its superiority in group fairness enhancement and fairness-utility tradeoff.
- 71
- StableMoFusion: Towards Robust and Efficient Diffusion-based Motion Generation Framework70
To eliminate footskate, the StableMoFusion model identifies foot-ground contact and correct foot motions along the denoising process, and is presented as a robust and efficient framework for human motion generation.
- 70
- Deep Convolutional Neural Networks with Merge-and-Run Mappings70
This work shows that the merge-and-run mapping is a linear idempotent function in which the transformation matrix is idem Potent, and thus improves information flow, making training easy.
- Fully Sparse Fusion for 3D Object Detection67
This paper's proposed detector integrates the well-studied 2D instance segmentation into the LiDAR side, which is parallel to the 3D instance segmentation part in the LiDAR-only baseline, and maintains full sparsity while overcoming the constraints associated with the LiDAR-only fully sparse detector.
- 64
- GAIA: A Transfer Learning System of Object Detection that Fits Your Needs60
GAIA is capable of providing powerful pre-trained weights, selecting models that conform to downstream demands such as latency constraints and specified data domains, and collecting relevant data for practitioners who have very few datapoints for their tasks.
- Employing Multi-estimations for Weakly-Supervised Semantic Segmentation60
This work proposes a novel approach to alleviate the inaccurate seed problem by leveraging the segmentation model’s robustness to learn from multiple seeds, and generates many different seeds for each image that are estimates of the underlying ground truth.
- Spatiotemporal distilled dense-connectivity network for video action recognition56
This work proposes a Spatiotemporal Distilled Dense-Connectivity Network (STDDCN) for video action recognition that implements both knowledge distillation and dense-connectivity (adapted from DenseNet) and achieves promising performances.
- Trim 3D Gaussian Splatting for Accurate Geometry Representation55
The experimental and theoretical analyses reveal that a relatively small Gaussian scale is a non-negligible factor in representing and optimizing the intricate details in representing and optimizing the intricate details, and the proposed TrimGS maintains relatively small Gaussian scales.
- Attention-Aware Sampling via Deep Reinforcement Learning for Action Recognition55
An attentionaware sampling method for action recognition, which aims to discard the irrelevant and misleading frames and preserve the most discriminative frames, which can be applied to different existing deep learning based action recognition models.
- 55
- Dither modulation of significant amplitude difference for wavelet based robust watermarking55
Experimental results demonstrate the superiority of the proposed wavelet quantization based method on robustness against content-preserving operations and incidental distortions such as JPEG compression, Gaussian noise.
- Self-Supervised Predictive Learning: A Negative-Free Method for Sound Source Localization in Visual Scenes54
This paper designs a three-stream network to elegantly associate sound source with two augmented views of one corresponding video frame, leading to semantically coherent similarities between audio and visual features, and introduces a novel predictive coding module for audio-visual feature alignment.
- 54
- 52
- Skeleton-Based Relational Modeling for Action Recognition.50
An Attentional Recurrent Relational Network-LSTM (ARRN-L STM) to simultaneously model spatial configurations and temporal dynamics in skeletons for action recognition and introduces an adaptive attentional module for focusing on potential discriminative parts of the skeleton towards a certain action.
- HelloBench: Evaluating Long Text Generation Capabilities of Large Language Models48
The Hierarchical Long Text Generation Benchmark (HelloBench), a comprehensive, in-the-wild, and open-ended benchmark to evaluate LLMs' performance in generating long text and the proposed HelloEval, a human-aligned evaluation method that significantly reduces the time and effort required for human evaluation while maintaining a high correlation with human evaluation.
- Once Detected, Never Lost: Surpassing Human Performance in Offline LiDAR based 3D Object Detection46
The proposed high-performance offline LiDAR-based 3D object detection method surpassing the human-level annotating accuracy and previous state-of-the-art methods in the highly competitive Waymo Open Dataset leaderboard without model ensemble.
- Super Sparse 3D Object Detection46
A fully sparse object detector termed FSD, built upon the general sparse voxel encoder and a novel sparse instance recognition (SIR) module, and a super sparse detector named FSD++, which first generates residual points, which indicate the point changes between consecutive frames.
- 46
- A Survey on Personalized Content Synthesis with Diffusion Models44
This paper provides a comprehensive survey of PCS, introducing the general frameworks of PCS research, which can be categorized into test-time fine-tuning (TTF) and pre-trained adaptation (PTA) approaches, and analyzes the strengths, limitations and key techniques of these methodologies.
- Blind Video Deflickering by Neural Filtering with a Flawed Atlas44
This work proposes a general flicker removal framework that only receives a single flickering video as input without additional guidance, and achieves satisfying deflickering performance and even outperforms baselines that use extra guidance on a public benchmark.
- 44
- 44
- 43
- A novel self-recovery fragile watermarking scheme based on dual-redundant-ring structure43
Experimental results demonstrate that the proposed algorithm is superior to the state of the art of tamper detection and tamper recovery and could keep high security strength.
- DexVLG: Dexterous Vision-Language-Grasp Model at Scale42
Extensive experiments demonstrate DexVLG's strong zero-shot generalization capabilities, achieving an over 76 % zero-shot execution success rate and state-of-the-art part-grasp accuracy in simulation, as well as successful part-aligned grasps on physical objects in real-world scenarios.
- 42
- Unsupervised Domain Adaptation with Background Shift Mitigating for Person Re-Identification42
A Densely Associated 2-Stream (DA-2S) network with an update strategy to best learn discriminative ID features from generated data that consider both human body information and also certain useful ID-related cues in the environment is proposed.
- 42
- SceneX: Procedural Controllable Large-Scale Scene Generation41
A large-scale scene generation framework, SceneX, is introduced, which can automatically produce high-quality procedural models according to designers' textual descriptions, and aims to generate executable actions for Blender to produce controllable and precise 3D assets guided by the user's instructions.
- 41
- 41
- 41
- 41
- Semi-supervised domain adaptation via Fredholm integral based kernel methods41
This paper proposes a novel Transfer Fredholm Multiple Kernel Learning (TFMKL) framework to suppress the noise for complex data distributions, and emphasizes the adaptability of TFMKL to different domain adaptation tasks due to its extension to different predictive models.
- Tamper detection and self-recovery of biometric images using salient region-based authentication watermarking scheme40
A novel scheme for tamper detection and self-recovery of biometric images using salient region-based authentication watermarking is proposed, which can detect tampered regions, and recover biometric data while keeping the recognition quality.
- Distractor-Aware Fast Tracking via Dynamic Convolutions and MOT Philosophy39
This work proposes a two-task tracking framework (named DMTrack), which utilizes two core components to achieve distractor-aware fast tracking via Dynamic convolutions (d-convs) and Multiple object tracking (MOT) philosophy, and constructs a lightweight one-shot detector using a novel dynamic convolutions generation method.
- Multi-task Layout Analysis for Historical Handwritten Documents Using Fully Convolutional Networks39
Experimental results on the public dataset DIVA-HisDB containing challenging medieval manuscripts demonstrate the effectiveness and superiority of the proposed multi-task layout analysis method.
- Integrating Both Visual and Audio Cues for Enhanced Video Caption38
This work proposes a dynamic multimodal feature fusion framework to deal with the part modalities missing case and demonstrates that even in the audio absence mode, results can still be obtained with the aid of the additional audio modality inference module.
- Context-aware co-supervision for accurate object detection37
This paper advocates the importance of equipping two-stage detectors with top-down signals, in order to which provides high-level contextual cues to complement low-level features in object detection.
- Efficient auto-refocusing for light field camera37
This paper presents an efficient ARF solution for light-field cameras based on modeling the refocusing point spread function (R-PSF), which is much more efficient than the current “rendering and selection” solutions which need to render a large number of refocused images.
- Secure multimodal biometric authentication with wavelet quantization based fingerprint watermarking37
A wavelet quantization based watermarking approach is proposed to adaptively distribute watermark energy on significant DWT coefficients of fingerprint images to guarantee the robustness and capacity of watermark while preserving the discriminating features of host fingerprints.
- 37
- Brain-inspired artificial intelligence research: A review35
The role of cognitive science in inspiring the development of the three mainstream academic branches of AI based on the three-layer framework proposed by David Marr is reviewed, and the limitations of the current development of AI are explored and analyzed.
- 35
- Unsupervised Object Detection with LiDAR Clues35
This paper exploits LiDAR clues to aid unsupervised object detection by exploiting the 3D scene structure, and identifies another major issue, seldom noticed by the community, that the long-tailed and open-ended (sub-)category distribution should be accommodated.
- Dynamic Multi-Task Learning with Convolutional Neural Network35
This paper considers the integration of CNN and multi-task learning in a novel way to further improve the performance of multiple related tasks and proposes a dynamic multi-task CNN model that outperforms traditional approaches.
- Air Quality Measurement Based on Double-Channel Convolutional Neural Network Ensemble Learning34
The double-channel weighted convolutional network (DCWCN) ensemble learning algorithm, which uses each channel to train different parts of the environment images for feature extraction and a feature weights self-learning method, which weights and concatenates the extracted feature vectors to measure the air quality.
- Protecting Biometric Templates Using Authentication Watermarking34
Experimental results show that the proposed scheme can detect the tampered region, and recover the biometric features while keeping recognizing quality.
- 33
- 32
- FrustumFormer: Adaptive Instance-aware Resampling for Multi-view 3D Detection32
A novel framework named FrustumFormer is proposed, which pays more attention to the features in instance regions via adaptive instance-aware resampling, and achieves a new state-of-the-art performance on the benchmark.
- LMR: A Large-Scale Multi-Reference Dataset for Reference-based Super-Resolution32
A new baseline method for multi-reference super-resolution is proposed: MRefSR, including a Multi-Reference Attention Module (MAM) for feature fusion of an arbitrary number of reference images, and a Spatial Aware Filtering Module (SAFM) for the fused feature selection.
- 32
- 32
- CityX: Controllable Procedural Content Generation for Unbounded 3D Cities31
The method, named CityX, demonstrates its superiority in creating diverse, controllable, and realistic 3D urban scenes that can be seamlessly deployed as a real-time simulator and an infinite data generator for embodied intelligence research.
- Random Shifting for CNN: a Solution to Reduce Information Loss in Down-Sampling Layers31
A novel random strategy is proposed to alleviate problems of poor generalization ability and information loss in deep convolutional neural networks by embedding random shifting in the down-sampling layers during the training process.
- 31
- Crowd density estimation based on statistical analysis of local intra-crowd motions for public area surveillance31
A novel accumulated mosaic image difference feature is proposed to represent these complicated random motion patterns in crowds for accurate foreground detection and is normalized based on the perspective distortion correction model to achieve a reasonable crowd density measurement for observed areas.
- 29
- Implementation of EEG Emotion Recognition System Based on Hierarchical Convolutional Neural Networks28
This paper introduces hierarchical convolutional neural networks (HCNN) to implement the EEG-based emotion classifier (positive, negative and neutral) in a movie-watching task and finds Beta and Gamma waves play the key role in emotion recognition.
- 27
- Uncertainty-optimized deep learning model for small-scale person re-identification27
This study considers the uncertainty of pedestrian representation for small-scale person re-identification and designs an improved Monte Carlo strategy that considers both the average distance and shortest distance for matching and ranking.
- Multi-block dependency based fragile watermarking scheme for fingerprint images protection27
The proposed multi-block dependency based fragile watermarking scheme can detect and localize the isolated-block tamper on fingerprint images with high detection probability and low false detection probability, but also enhances the systematic security obviously.
- 26
- POD: Practical Object Detection With Scale-Sensitive Network26
This paper proposes a scale decomposition method that transfers the robust fractional scale into combinations of fixed integral scales for each convolution filter, which exploit the dilated convolution.
- A Curriculum-Style Self-Training Approach for Source-Free Semantic Segmentation25
A curriculum-style entropy minimization method is introduced to explore the implicit knowledge from the source model, which fits the trained source model to the target data using certain information from easy-to-hard predictions, which could act as a standard post-processing method for the domain adaptation field.
- MemoNav: Working Memory Model for Visual Navigation25
MemoNav is presented, a novel memory model for image-goal navigation, which utilizes a working memory-inspired pipeline to improve navigation performance and significantly outperforms previous methods across all difficulty levels in both Gibson and Matterport3D scenes.
- 25
- Video-Based Face Recognition: State of the Art25
This paper divides the existing video based face recognition approaches into two categories: videoimage based methods and video-video based methods, which are surveyed and analyzed in this paper.
- 24
- DDG-Net: Discriminability-Driven Graph Network for Weakly-supervised Temporal Action Localization23
Discriminability-Driven Graph Network (DDG-Net) is proposed, which explicitly models ambiguous snippets and discriminative snippets with well-designed connections, preventing the transmission of ambiguous information and enhancing the discriminability of snippet-level representations.
- 22
- 22
- Generative adversarial dehaze mapping nets22
A Generative Adversarial Dehaze Mapping Nets (GADMN) to estimate a medium transmission for an input hazy image, which outperforms state-of-the-art methods, and exhibits better robustness and adaptability.
- 21
- Robust Depth Enhancement via Polarization Prompt Fusion Tuning20
This work presents a general framework that leverages polarization imaging to improve inaccurate depth measurements from various depth sensors and proposes a Polarization Prompt Fusion Tuning (PPFT) strategy to effectively utilize RGB-based models pre-trained on large-scale datasets.
- From Individual to Whole: Reducing Intra-class Variance by Feature Aggregation20
Inspired by the equivalence between the normalized cut and random walk, a feature aggregation scheme using transition probabilities as guidance is proposed that is on par with, or even better than current state-of-the-arts in both tasks.
- HP-Capsule: Unsupervised Face Part Discovery by Hierarchical Parsing Capsule Network20
Hierarchical Parsing Capsule Network (HP-Capsule) is proposed, which extends the application of capsule networks from digits to human faces and takes a step forward to show how the neural networks understand homologous objects without human intervention.
- MTU-Bench: A Multi-granularity Tool-Use Benchmark for Large Language Models19
A multi-granularity tool-use benchmark for large language models called MTU-Bench, which is collected by transforming existing high-quality datasets to simulate real-world tool usage scenarios, and an instruction dataset called MTU-Instruct data to enhance the tool-use abilities of existing LLMs.
- 19
- Extracting Semantic Knowledge From GANs With Unsupervised Learning19
KLiSH succeeds in extracting fine-grained semantics of GANs trained on datasets of various objects, e.g., car, portrait, animals, and so on, and it is found that GAN's features can be well clustered with the linear separability assumption.
- 19
- Conditional Expression Synthesis with Face Parsing Transformation19
A Couple-Agent Face Parsing based Generative Adversarial Network (CAFP-GAN) that unites the knowledge of facial semantic regions and controllable expression signals is introduced that has the compelling ability on continuous expression synthesis.
- 18
- Toward Practical Weakly Supervised Semantic Segmentation via Point-Level Supervision18
This work proposes a superpixel augmented pseudo-mask generation strategy and a class-aware contrastive learning approach, which manages to recover reliable dense constraints and apply them both to the segmentation models’ final prediction and the intermediate features.
- 18
- 17
- 17
- 17
- SAS: Painting Detection and Recognition via Smart Art System With Mobile Devices16
This paper proposes a Smart Art System (SAS) with mobile devices that achieves the highest performance and better real-time performance, and evaluates the method and compares with the relevant algorithms.
- View independent object classification by exploring scene consistency information for traffic scene surveillance16
Two solutions are given to achieve automatic object classification based on simple motion and shape features on the 2D image plane, both of which are free of large database collection and manually labeling.
- 16
- 16
- From Code Foundation Models to Agents and Applications: A Comprehensive Survey and Practical Guide to Code Intelligence15
This work provides a comprehensive synthesis and practical guide about code LLMs, systematically examining the complete model life cycle from data curation to post-training through advanced prompting paradigms, code pre-training, supervised fine-tuning, reinforcement learning, and autonomous coding agents.
- Enhancing Visual Continual Learning with Language-Guided Supervision15
This paper revisit the role of the classifier head within the CL paradigm and replace the classifier with semantic knowledge from pretrained language models (PLMs) to generate semantic targets for each class, which are frozen and serve as supervision signals during training.
- Informative Data Mining for One-shot Cross-Domain Semantic Segmentation15
This work proposes a novel framework called Informative Data Mining (IDM) that enables efficient one-shot domain adaptation for semantic segmentation and achieves a new state-of-the-art one-shot performance of 56.7%/55.4% on the GTA5/SYNTHIA to Cityscapes adaptation tasks, respectively.
- Real-Aug: Realistic Scene Synthesis for LiDAR Augmentation in 3D Object Detection15
This work systematically study the synthesis-based LiDAR data augmentation approach (so-called GT-Aug) which offers maxium controllability over generated data samples and proposes Real-Aug, a synthesis- based augmentation method which prioritizes on generating realisticLiDAR scans.
- 15
- 15
- 14
- 14
- 14
- 14
- 14
- GRAMO: geometric resampling augmentation for monocular 3D object detection13
Two geometric-level data augmentation operators named Geometric-Copy-Paste (Geo-CP) and Geometric-Crop-Shrink (Geo-CS) are proposed, which introduce geometric consistency based on the principle of perspective projection, complementing the options available for data augmentation in monocular 3D.
- Fast Hyperspectral Image Classification with Strong Noise Robustness Based on Minimum Noise Fraction13
A fast hyperspectral image classification algorithm with strong noise robustness is proposed in this paper, aiming at the hyperspectral image classification problems under noise interference by using a Minimum Noise Fraction as dimensionality reduction module for hyperspectral image input data.
- MixSup: Mixed-grained Supervision for Label-efficient LiDAR-based 3D Object Detection13
This work proposes MixSup, a more practical paradigm simultaneously utilizing massive cheap coarse labels and a limited number of accurate labels for Mixed-grained Supervision, and proposes PointSAM based on the Segment Anything Model for automated coarse labeling, further reducing the annotation burden.
- 13
- Self-Guided Hard Negative Generation for Unsupervised Person Re-Identification13
A joint framework is developed which incorporates a hard negative generation network (HNGN) and a re-ID network which is alternately trained in an adversarial manner to improve each other, where the reID network guides H NGN to generate challenging data and HNGN enforces the re-IDs to enhance discrimination ability.
- Face synthesis from low-resolution near-infrared to high-resolution visual light spectrum based on tensor analysis13
Compared to the two state-of-the-art algorithms, KNN and LBP-KNN, this approach shows better robustness to moderate pose and expression variations, and outstanding efficiency in dealing with images of poor quality and low resolutions.
- 13
- 13
- 13
- Learnable Graph Matching: A Practical Paradigm for Data Association12
To make optimization end-to-end differentiable, the original graph matching problem into continuous quadratic programming and then incorporate training into a deep graph neural network with KKT conditions and implicit function theorem is relaxed.
- 12
- 12
- DiffusePast: Diffusion-based Generative Replay for Class Incremental Semantic Segmentation12
DiffusePast introduces a dual-generator paradigm, which focuses on generating old class images that align with the distribution of downstream datasets while preserving the structure and layout of the original images, enabling more precise masks.
- Intrinsic Physical Concepts Discovery with Object-Centric Predictive Models12
This paper introduces the PHYsi-cal Concepts Inference NEtwork (PHYCINE), a system that infers physical concepts in different abstract levels with-out supervision and shows that object representations containing the discovered physical concepts variables could help achieve better performance in causal reasoning tasks, i.e., ComPhy.
- 12
- 12
- CodeCriticBench: A Holistic Code Critique Benchmark for Large Language Models11
A holistic code critique benchmark for LLMs called CodeCriticBench is introduced, which includes two mainstream code tasks (i.e., code generation and code QA) with different difficulties and the evaluation protocols include basic critique evaluation and advanced critique evaluation for different characteristics.
- MaterialSeg3D: Segmenting Dense Materials from 2D Priors for 3D Assets11
This work proposes MaterialSeg3D, a 3D asset material generation framework to infer underlying material from the 2D semantic prior, and devise a mechanism to parse material in 3D space.
- Delving into the Effectiveness of Receptive Fields: Learning Scale-Transferrable Architectures for Practical Object Detection11
A scale-transferrable architecture for practical object detection based on the analysis of the connection between dilation rate and effective receptive field and a scale decomposition method that transfers the robust fractional scale into the combination of fixed integral scales for each convolution filter, which exploits the dilated convolution.
- DATA: Domain-Aware and Task-Aware Self-supervised Learning11
This paper presents DATA, a simple yet effective NAS approach specialized for SSL that provides Domain-Aware and Task- aware pre-training and proposes a flexible searching mechanism compatible with SSL that enables finding networks of different computation costs, for various downstream vision tasks and data domains without explicit metric provided.
- 11
- 11
- 11
- OOD-HOI: Text-Driven 3D Whole-Body Human-Object Interactions Generation Beyond Training Domains10
This paper proposes OOD-HOI, a text-driven framework for generating whole-body human-object interactions that generalize well to new objects and actions that integrates a dual-branch reciprocal diffusion model, a contact-guided interaction refiner, and a dynamic adaptation mechanism which includes semantic adjustment and geometry deformation to improve robustness.
- GPT4SGG: Synthesizing Scene Graphs from Holistic and Region-specific Narratives10
Experimental results demonstrate \textit{GPT4SGG} significantly improves the performance of SGG models trained on image-caption data, in which the ambiguity issue and long-tail bias have been well-handled with more accurate and comprehensive scene graphs.
- Graphics Capsule: Learning Hierarchical 3D Face Representations from 2D Images10
An Inverse Graphics Capsule Network (IGC-Net) is proposed to learn the hierarchical 3D face representations from large-scale unlabeled images and reveals how the neural networks, oriented at visual perception, understand faces as a hierarchy of 3D models.
- Improving Image Segmentation with Boundary Patch Refinement10
This work proposes a conceptually simple yet effective post-processing refinement framework, termed BPR, to improve the boundary quality of the prediction of any image segmentation model and extracts and refine a series of small boundary patches along the predicted boundaries.
- Toward few-shot domain adaptation with perturbation-invariant representation and transferable prototypes10
This work designs a data perturbation strategy and proposes a transferable prototype module to bridge the domain gap based on the source data and few-shot targets and can perform on par with the fully supervised models to some extent.
- 10
- Biologically inspired visual computing: the state of the art10
This work will present a survey of the current work on biologically inspired visual computing, and hope to offer some new avenues for rethinking visual computing and designing novel neural network architectures.
- Advances in Brain Inspired Cognitive Systems10
An improved recurrent network and its electronic implementation for online equality-constrained quadratic programming with linear equality constraints is proposed and developed.
- Automatic object classification using motion blob based local feature fusion for traffic scene surveillance10
This paper evaluates the performance of local feature descriptors for automatic object classification in traffic scenes with robustness to noise, variance of view angles, lighting conditions, and environments, and demonstrates the efficiency and effectiveness of this strategy.
- 10
- 9
- 9
- Visually Guided Sound Source Separation With Audio-Visual Predictive Coding9
This article presents a novel approach, dubbed audio- visual predictive coding (AVPC), to tackle the task of visually guided sound source separation in a parameter-efficient and more effective manner, and develops a valid self-supervised learning strategy for AVPC via copredicting two audio-visual representations of the same sound source.
- Stereo Depth Estimation with Echoes9
An end-to-end framework named StereoEchoes for stereo depth estimation with echoes is proposed, designed to transfer the complementary knowledge of the audio modality to the visual modality at feature level and yield pixel-wise confidence for multimodal depth fusion at output space.
- A Primal Neural Network for Online Equality-Constrained Quadratic Programming9
An improved primal recurrent neural network is proposed and its electronic implementation and theoretical analysis are proposed, which verify the effectiveness of the proposed model for online equality-constrained quadratic programming.
- End-to-End Lifelong Learning: a Framework to Achieve Plasticities of both the Feature and Classifier Constructions9
This paper aims at exploring an end-to-end lifelong learning framework to achieve plasticities of both the feature and classifier constructions and has achieved better results than PCANet and KmeansNet on minist and Caltech101 datasets respectively.
- 9
- 9
- REPRESENTING 3D FACE FROM POINT CLOUD TO FACE-ALIGNED SPHERICAL DEPTH MAP9
A novel representation of 3D face shape which is a key step for feature extraction and face recognition, and face alignment based on Spherical Depth Map (SDM), which can be adopted to many applications.
- 9
- Aligning Instruction Tuning with Pre-training8
Aligning Instruction Tuning with Pre-training (AITP) is proposed, a method that bridges this gap by identifying coverage shortfalls in instruction-tuning datasets and rewriting underrepresented pre-training data into high-quality instruction-response pairs and enriches dataset diversity while preserving task-specific objectives.
- Dual-Image Enhanced CLIP for Zero-Shot Anomaly Detection8
This work introduces a Dual-Image Enhanced CLIP approach, leveraging a joint vision-language scoring system that markedly enhanced both anomaly classification and localization performances and strengthens the model with a test-time adaptation module that incorporates synthesized anomalies to refine localization capabilities.
- 8
- 8
- Spectral attribute learning for visual regression8
An approach to automatically discover spectral attributes which avoids manual work required for defining hand-crafted attribute representations is proposed and this spectral attribute visual regression achieves state-of-the-art accuracy in multiple applications.
- Incremental PCANet: A Lifelong Learning Framework to Achieve the Plasticity of both Feature and Classifier Constructions8
The proposed model mainly comprises of three parts: Gabor filters followed by maxpooling layer offering shift and scale tolerance to input samples, cascade incremental PCA to achieve the plasticity of feature extraction and incremental SVM to pursue Plasticity of classifier construction.
- 8
- Practical Continual Forgetting for Pre-Trained Vision Models7
This work introduces Low-Rank Adaptation (LoRA) modules to fine-tune the Feed-Forward Network layers in Transformer blocks for each forgetting task independently, and adopts a simple group sparse regularization, enabling automatic selection of specific LoRA groups and zeroing out the others.
- LiDAR-camera Cooperative Semantic Segmentation7
This work proposes a unified multimodal cooperative segmentation workflow that performs LiDAR-based 3D semantic segmentation and image-based 2D semantic segmentation with promising results on two public benchmarks: the SemanticKITTI dataset and the Waymo Open dataset.
- 7
- A Hierarchical Absolute Visual Localization System for Low-Altitude Drones in GNSS-Denied Environments7
Results indicate that the proposed technology can achieve high-precision and robust absolute visual localization without the need for initial conditions, highlighting its potential for application in GNSS-denied environments.
- 7
- 7
- A Survey of Deep Visual Cross-Domain Few-Shot Learning7
Through this review, this survey aims to provide comprehensive guidance on CDFS research, enabling researchers to gain insight into the state-of-the-art while allowing them to build upon existing solutions to develop their own CDFS models.
- 7
- Training Better CNNs Requires to Rethink ReLU.7
The proportional module with N:M (N$>$M) convolution and ReLU ratio can help networks acquire the better performance, through the analysis of a simple Leaky ReLU model and the experimental results verify that the superiority of the proportional module is found.
- Facial Age Estimation Using Robust Label Distribution7
A novel concept of robust label distribution for scalar-valued labels is introduced, which is designed to encode the age scalars into label distribution matrices, i.e. two-dimensional Gaussian distributions along age classes and age groups respectively.
- A Robust Watermarking Scheme Based on Dual Quantization of Wavelet Significant Difference7
A blind robust watermarking algorithm based on the quantization of significant DWT coefficients that demonstrates the superiority of this scheme on robustness against content-preserving operations and incidental distortions such as JPEG compression, Gaussian noise.
- 6
- Robust pose estimation for non-cooperative space objects based on multichannel matching method6
Extensive experimental results indicate that the proposed method demonstrates robust performance across various challenging conditions, including different object types, diverse illumination scenarios, varying rotation rates, and different image resolutions.
- Enhancing Sound Source Localization via False Negative Elimination6
A novel audio-visual learning framework which is instantiated with two individual learning schemes: self-supervised predictive learning (SSPL) and semantic-aware contrastive learning (SACL), designed to compact visual features and remove false negatives.
- Generative Active Learning for Image Synthesis Personalization6
A direction-based uncertainty sampling strategy to enable generative active learning and tackle the exploitation-exploration dilemma is proposed and it is demonstrated that an open-source model can achieve superior performance compared to closed-source models developed by large companies, such as Google's StyleDrop.
- 6
- 6
- Reusable Architecture Growth for Continual Stereo Matching6
This work proposes to perform continual stereo matching where a model is tasked to continually learn new scenes, overcome forgetting previously learned scenes, and continuously predict disparities at inference, by introducing a Reusable Architecture Growth (RAG) framework.
- Bootstrap Masked Visual Modeling via Hard Patches Mining6
This work proposes Hard Patches Mining (HPM), predicting patch-wise losses and subsequently determining where to mask, and introduces an auxiliary loss predictor, which is trained with a relative objective to prevent overfitting to exact loss values.
- 6
- Deconfounding Physical Dynamics with Global Causal Relation and Confounder Transmission for Counterfactual Prediction6
A neural network framework combining Global Causal Relation Attention (GCRA) and Confounder Transmission Structure (CTS) is proposed that outperforms the state-of-the-art method on various benchmarks by fully utilizing the constraints of confounders.
- Pattern Recognition and Information Forensics6
An overview of recent research results that have used Ocean Networks Canada imagery combined with oceanographic sensor data to study seafloor ecosystem dynamics and a review of a recent study that compared crowd sourcing results with expert annotation and computer algorithms to quantify the abundance of sablefish in a submarine canyon habitat.
- A Secure Semi-fragile Self-recoverable Watermarking Algorithm Using Group-Based Wavelet Quantization6
A secure semi-fragile self-recoverable watermarking algorithm using group-based wavelet quantization that embeds watermark in largest coefficient inside a sub-group based on significant difference parity quantization (SDPQ) method.
- Spam Short Messages Detection via Mining Social Networks6
This work proposes a novel approach for spam messages detection based on mining under a more robust structure: the social network constructed with SMS, which includes several features, including static features, dynamic features and graph features.
- 6
- 6
- FlexDrive: Toward Trajectory Flexibility in Driving Scene Gaussian Splatting Reconstruction and Rendering5
An Inverse View Warping technique is introduced to create compact and high-quality images as supervision for the reconstruction of the out-of-path views, enabling high-quality rendering results for those views, where the method outperforms previous methods by a significant margin.
- Joint optimization of sensor placement and sparse pressure field reconstruction with a two-stage framework for limited data5
The proposed PROSNet (Pressure-field Reconstruction with Optimal Sensor placement), a neural network framework addressing both challenges of sparse pressure field reconstruction and aerodynamic design, includes a differentiable node selection module that learns optimal sensor placements and adopts a two-stage training strategy that encourages exploration of diverse sensor combinations.
- Unsupervised Multimodal UAV Image Registration via Style Transfer and Cascade Network5
The proposed approach provides a robust solution for cross-modal image registration in dynamic UAV environments, offering significant implications for downstream applications such as target detection, tracking, and surveillance.
- Seek for Incantations: Towards Accurate Text-to-Image Diffusion Synthesis through Prompt Engineering5
This paper proposes a framework to learn the proper textual descriptions for diffusion models through prompt learning by utilizing the quality guidance and the semantic guidance derived from the pre-trained diffusion model to improve the matches between the input text and the generated images.
- 5
- 5
- 5
- An Improved Recurrent Network for Online Equality-Constrained Quadratic Programming5
An improved recurrent network and its electronic implementation for online equality-constrained quadratic programming with linear equality constraints is proposed and developed.
- Incremental learning patch-based bag of facial words representation for face recognition in videos5
An incremental learning approach to video-based face recognition which efficiently exploits the spatiotemporal information in videos is introduced and achieves a 100 % verification rate on the Honda/UCSD database and 82 % on the YouTube datebase.
- 5
- 5
- 4
- MemoNav: Selecting Informative Memories for Visual Navigation4
The experimental results show that the MemoNav outperforms the SoTA methods by a large margin with a smaller fraction of navigation history, and empirically shows that the model is less likely to be trapped in a deadlock, which further validates that thememoNav improves the agent's navigation efficiency by reducing redundant steps.
- 4
- Towards Independent Color Space Selection for Human Skin Detection4
A novel skin color detection method based on the flexible neural tree, which can identify the important components of color spaces automatically and achieves higher accuracy and lower false positive rate than state of the art methods on Compaq and ECU data set is proposed.
- 4
- 3
- Adaptive Missile Avoidance Algorithm for UAV Based on Multi-Head Attention Mechanism and Dual Population Confrontation Game3
This work introduces a risk-sensitive PPO algorithm and proposes a training framework incorporating multi-head attention mechanisms and dual-population adversarial training that demonstrates superior training efficiency and generates evasion policies with better adaptability to different missile types.
- 3
- 3
- 3
- Fully Data-Driven Pseudo Label Estimation for Pointly-Supervised Panoptic Segmentation3
This work estimates pseudo labels with a fully data-driven pseudo label branch, optimized by point labels end-to-end and predicts more accurate pseudo labels than previous methods.
- Interact with Open Scenes: A Life-long Evolution Framework for Interactive Segmentation Models3
A life-long evolution framework for interactive models is proposed, which provides a possible solution for dealing with dynamic target scenes with one single model and demonstrates the effectiveness of the framework in solving new target datasets and maintaining performance on previous scenes at the same time.
- 3
- 3
- Biometric information hiding: Promoting multimedia security with content and identity authentication3
- Applying local Gabor ternary pattern for video-based illumination variable face recognition3
A new approach which achieved illumination variable face recognition by combining Gabor filters with LTP operator is proposed, and the experimental results compared with the published results verify the validity of the proposed method.
- 3
- EmbodiedCoder: Parameterized Embodied Mobile Manipulation via Modern Coding Model2
EmbodiedCoder is introduced, a training-free framework for open-world mobile robot manipulation that leverages coding models to directly generate executable robot trajectories and demonstrates an interpretable approach for bridging high-level reasoning and low-level control, moving beyond fixed primitives toward versatile robot intelligence.
- Exploiting Diffusion Priors for Generalizable Few-Shot Satellite Image Semantic Segmentation2
A novel diffusion-based framework for few-shot satellite segmentation, named DiffSatSeg, which leverages the powerful compositional generalization capability of diffusion models to address the challenges inherent in satellite segmentation tasks and enables reference-based segmentation, which holds great potential for practical deployment and real-world applications.
- 2
- 2
- 2
- 2
- 2
- 2
- Biomechanics of Negative-Pressure-Assisted Liposuction and Their Influence on Fat Regeneration2
This review innovatively employs biomechanical knowledge to link clinical instruments, laboratory experiments, and animal studies together, providing alternative strategies for plastic surgeons to choose harvesting cannulas during fat grafting.
- Tracking Objects with 3D Representation from Videos2
This paper proposes a new 2D MOT paradigm, called P3DTrack, which utilizes the 3D object representation to separate each object in the feature space and achieves new state-of-the-art performance on the large-scale Waymo Open Dataset.
- 2
- 2
- Human-Like Delicate Region Erasing Strategy for Weakly Supervised Detection2
A novel human-like active searching strategy that recurrently ignores the background and discovers class-specific objects by erasing undesired pixels from the image and is the first attempt to apply reinforcement learning to address weakly supervised localization with only image-level labels.
- 2
- Visual Information Processing Mechanism Revealed by fMRI Data2
The functional Magnetic Resonance Imaging (fMRI) data of both the ventral pathway and the dorsal pathway on the visual cortex in a classification task was analyzed and found that the classification performance improved hierarchically from lower- level regions to higher-level regions in both pathways, which partly verified the visual pathway theory proposed in cognitive neuroscience.
- 2
- 2
- 2
- On-line signature verification based on spatio-temporal correlation2
A novel strategy is proposed in this paper by combining texture based image analysis and spatio-temporal representation to address the problem of on-line signature representation and verification.
- Model-Based Multi-view Face Construction and Recognition in Videos2
A novel method to model and recognize multi-view faces in video sequences by using faces' paths in different view and feature points obtained from models is presented, synthesized by reconstruction and stitching the paths together.
- 2
- Investigation of the Robustness and Transferability of Adversarial Patches in Multi-View Infrared Target Detection1
Compared with existing adversarial attack methods such as advYOLO Patch and QR Attack, GADP exhibits superior transferability and attack performance, reducing the Average Precision of multiple detection models to around 40%.
- 1
- 1
- 1
- UniSPAC: A Unified Segmentation Framework for Proofreading and Annotation in Connectomics1
A Unified Segmentation framework for Proofreading and Annotation in Connectomics (UniSPAC) is introduced by providing the interactive segmentation model in 2D-level and the neuron tracing model in 3D-level, combining segmentation and proofreading in a single framework.
- 1
- Robust Feature Rectification of Pretrained Vision Models for Object Recognition1
ROFER first estimates the type and intensity of the degradation that corrupts the image features, then leverages a Fully Convolutional Network (FCN) to rectify the features from the degradation by pulling them back to clear features.
- Visual Commonsense based Heterogeneous Graph Contrastive Learning1
This work incorporates the visual commonsense information and proposes a heterogeneous graph contrastive learning method to better finish the visual reasoning task, designed as a plug-and-play way that can be quickly and easily combined with a wide range of representative methods.
- 1
- Manual-Label Free 3D Detection via An Open-Source Simulator1
A manual-label free 3D detection algorithm that leverages the CARLA simulator to generate a large amount of self-labeled training samples and introduces a novel Domain Adaptive VoxelNet (DA-VoxelNet) that can cross the distribution gap from the synthetic data to the real scenario.
- On the role of sparsity in feature selection and an innovative method LRMI1
This paper proposes a novel feature selection algorithm, i.e. the Laplacian regularization based on mutual information (LRMI), which realizes the minimization of the redundancy in a new way, and incorporates the l2 norm to achieve automatic grouping.
- 1
- 1
- Active Learning for Transferrable Object Classification in Cross-View Traffic Scene Surveillance1
This paper proposes an active transfer learning method to decrease manual labeling work further for target scene traffic object classification and results demonstrate the effectiveness and convenience of this approach.
- Incremental Learning of Patch-Based Bag of Facial Words Representation for Online Face Recognition in Videos1
This paper introduces an incremental learning approach to video-based face recognition, which efficiently exploits the spatiotemporal information in videos and achieves a 100% recognition rate performed on the Honda/UCSD database, and gives near realtime feedback.
- View Independent Object Classification Based on Automated Ground Plane Rectification for Traffic Scene Surveillance1
An online learning framework is described to achieve automatic object classification based on rectified 2D object features to classify moving objects of traffic scene surveillance videos into pedestrians, bicycles and vehicles.
- –
- –
- –
- –
- –
- M2RC-EVAL: Massively Multilingual Repository-level Code Completion Evaluation–
A massively multilingual repository-level code completion benchmark covering 18 programming languages, and two types of fine-grained annotations on different completion scenarios are provided, to facilitate the research of code LLMs in multilingual scenarios.
- –
- A real-time and general method for converting offline skeleton-based action recognition to online ones–
This paper points out that to convert offline methods to online ones, the key is removing outdated frame features and fusing new frame features, and proposes a general and simple model called Encode One Frame (EOF), which achieves feature removal and fusion by a correlation matrix and the guidance of a teacher model.
- Steel-LLM:From Scratch to Open Source -- A Personal Journey in Building a Chinese-Centric LLM–
A comprehensive summary of the Steel-LLM project's key contributions, including data collection, model design, training methodologies, and the challenges encountered along the way are provided, offering a valuable resource for researchers and practitioners looking to develop their own LLMs.
- –
- –
- –
- –
- STODINE: Decompose video to Object-centric Spatial-Temporal Slots for physical reasoning–
A new unsupervised framework, STODINE (Spatial-Temporal Object Decomposition Inference Network), is introduced, which decomposes spatial slots from frames using amortized variational inference and distills temporal slots from a sequence of spatial slots through a multi-scale distillator.
- –
- Weakly-Supervised Object Localization by Cutting Background with Deep Reinforcement Learning–
This work proposes a novel weakly-supervised localization method of cutting background of an object iteratively to achieve object localization with deep reinforcement learning, and may be the first attempt to applyDeep reinforcement learning to weakly -supervised object localization.
- –
- Special Issue of BICS 2016–
This special issue aims to report new advances since BICS 2016, by including expanded versions of selected conference papers and also new contributions in brain information processing, braininspired cognitive models, and algorithms for decision, learning, vision, and applications.
- View Decomposition and Adversarial for Semantic Segmentation–
A novel discriminator module called Multi-View Decomposition is introduced which transforms the discriminator role from general teacher to specific adversary and separates single sample into a series of class inter-independent streams and extracts corresponding features from current mask.
- –
- Enhancing Human Pose Estimation with Temporal Clues–
This paper proposes a method and augmented Pose-NMS to process the human pose estimation in the consecutive frames based on a reasonable assumption and obtains the competitive result to the state-of-art.
- Tracking and recognition face in videos with incremental local sparse representation model–
A robust face tracking algorithm is proposed via employing local sparse appearance and covariance pooling method and a novel template update strategy, which adapts the template to appearance changes and reduces the influence of occlusion and illumination variation produces a robust video-based face tracking and recognition with desirable performance.
- Efficient Human Parsing Based on Sketch Representation–
An efficient human parsing method which estimates human body poses from 2D images by proposing an edge sketch representation, which enhance critical information for pose estimation and prune the redundant.
- –
- An Unsupervised Real-Time Tracking and Recognition Framework in Videos–
The framework is capable of tracking the face and automatically grouping a serial faces for a single human-being object in an unlabeled video robustly and is validated both on several videos collected in unconstrained condition and on Honda/UCSD database.
- –
- A Local Texture-Constrained Super-Resolution Method–
Through the learned low/high-resolution patches from training images, the intended high resolution patches are reconstructed using neighbor embedding method to propose a local texture constrained super-resolution method for the reconstruction of high-resolution image.
- –
- –
- Watermarking algorithm in multiple-parameter discrete fractional Fourier transform domain–
An image watermarking algorithm based on multiple-parameter discrete fractional Fourier transform(MPDFRFT) has better robustness and security than previous proposed algorithms.
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-14. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.