Pan Zhou lists you as their PhD advisor. Claim this profile to confirm it and keep the rest of your record right.
Claim this profileAcademic lineage
View as a treeStudents and postdocs2
Possible advisorsa guess from early papers, not confirmed
- Shuicheng YanSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works487 from public data
- Coordinate Attention for Efficient Mobile Network Design6,437
This paper proposes a novel attention mechanism for mobile networks by embedding positional information into channel attention, which it is called "coordinate attention", which factorizes channel attention into two 1D feature encoding processes that aggregate features along the two spatial directions.
- Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet2,580
A new Tokens-To-Token Vision Transformer (T2T-VTT), which incorporates an efficient backbone with a deep-narrow structure for vision transformer motivated by CNN architecture design after empirical study and reduces the parameter count and MACs of vanilla ViT by half.
- Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data2,221
This work designs a data engine to collect and automatically annotate large-scale unlabeled data and develops an auxiliary supervision to enforce the model to inherit rich semantic priors from pre-trained encoders, which results in a better depth-conditioned ControlNet.
- Return of Frustratingly Easy Domain Adaptation2,159
This work proposes a simple, effective, and efficient method for unsupervised domain adaptation called CORrelation ALignment (CORAL), which minimizes domain shift by aligning the second-order statistics of source and target distributions, without requiring any target labels.
- PANet: Few-Shot Image Semantic Segmentation With Prototype Alignment1,540
This paper tackles the challenging few-shot segmentation problem from a metric learning perspective and presents PANet, a novel prototype alignment network to better utilize the information of the support set to better generalize to unseen object categories.
- MetaFormer is Actually What You Need for Vision1,462
It is argued that MetaFormer is the key player in achieving superior results for recent transformer and MLP-like models on vision tasks, and calls for more future research dedicated to improving MetaFormer instead of focusing on the token mixer modules.
- Deep Joint Rain Detection and Removal from a Single Image1,331
A recurrent rain detection and removal network that removes rain streaks and clears up the rain accumulation iteratively and progressively is proposed and a new contextualized dilated network is developed to exploit regional contextual information and to produce better representations for rain detection.
- Tensor Robust Principal Component Analysis with a New Tensor Nuclear Norm1,142
This paper considers the Tensor Robust Principal Component Analysis (TRPCA) problem, which aims to exactly recover the low-rank and sparse components from their sum, and proposes a model based on the recently proposed tensor-tensor product, which includes matrix RPCA as a special case.
- A Simple Pooling-Based Design for Real-Time Salient Object Detection1,100
This work solves the problem of salient object detection by investigating how to expand the role of pooling in convolutional neural networks by building a global guidance module (GGM) and designing a feature aggregation module (FAM) to make the coarse-level semantic information well fused with the fine-level features from the top-down path- way.
- Deep Long-Tailed Learning: A Survey1,013
This article groups existing deep long-tailed learning studies into three main categories (i.e., class re-balancing, information augmentation and module improvement), and examines several state-of-the-art methods following this taxonomy in detail.
- Perceptual Generative Adversarial Networks for Small Object Detection913
This work addresses the small object detection problem by developing a single architecture that internally lifts representations of small objects to super-resolved ones, achieving similar characteristics as large objects and thus more discriminative for detection.
- Object Region Mining with Adversarial Erasing: A Simple Classification to Semantic Segmentation Approach881
This work investigates a principle way to progressively mine discriminative object regions using classification networks to address the weakly-supervised semantic segmentation problems and proposes a new adversarial erasing approach for localizing and expanding object regions progressively.
- Few-Shot Object Detection via Feature Reweighting879
This work develops a few-shot object detector that can learn to detect novel objects from only a few annotated examples, using a meta feature learner and a reweighting module within a one-stage detection architecture.
- Scale-aware Fast R-CNN for Pedestrian Detection843
This paper argues that the issue of large variance in instance scales, which results in undesirable large intracategory variance in features, may severely hurt the performance of modern object instance detection methods, can be substantially alleviated by the divide-and-conquer philosophy.
- Strip Pooling: Rethinking Spatial Pooling for Scene Parsing820
Beyond conventional spatial pooling that usually has a regular shape of NxN, a new pooling strategy is introduced, called strip pooling, which considers a long but narrow kernel, i.e., 1xN or Nx1.
- Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks With Octave Convolution735
This work proposes to factorize the mixed feature maps by their frequencies, and design a novel Octave Convolution (OctConv) operation to store and process feature maps that vary spatially “slower” at a lower spatial resolution reducing both memory and computation cost.
- 676
- Revisiting Dilated Convolution: A Simple Approach for Weakly- and Semi-Supervised Semantic Segmentation627
It is found that varying dilation rates can effectively enlarge the receptive fields of convolutional kernels and more importantly transfer the surrounding discriminative information to non-discriminative object regions, promoting the emergence of these regions in the object localization maps.
- Adversarial Complementary Learning for Weakly Supervised Object Localization626
This work mathematically proves that class localization maps can be obtained by directly selecting the class-specific feature maps of the last convolutional layer, which paves a simple way to identify object regions and presents a simple network architecture including two parallel-classifiers for object localization.
- STC: A Simple to Complex Framework for Weakly-Supervised Semantic Segmentation588
A simple to complex (STC) framework in which only image-level annotations are utilized to learn DCNNs for semantic segmentation, which demonstrates the superiority of the proposed STC framework compared with other state-of-the-arts frameworks.
- 586
- Natural Language Object Retrieval579
Experimental results demonstrate that the SCRC model effectively utilizes both local and global information, outperforming previous baseline methods significantly on different datasets and scenarios, and can exploit large scale vision and language datasets for knowledge transfer.
- Voxel Transformer for 3D Object Detection574
Voxel Transformer is presented, a novel and effective voxel-based Transformer backbone for 3D object detection from point clouds that shows consistent improvement over the convolutional baselines while maintaining computational efficiency on the KITTI dataset and the Waymo Open dataset.
- Tensor Robust Principal Component Analysis: Exact Recovery of Corrupted Low-Rank Tensors via Convex Optimization541
This work proves that under certain suitable assumptions, it can recover both the low-rank and the sparse components exactly by simply solving a convex program whose objective is a weighted combination of the tensor nuclear norm and the l1-norm.
- Correlation Alignment for Unsupervised Domain Adaptation534
This chapter describes a solution that applies a linear transformation to source features to align them with target features before classifier training, and proposes to equivalently apply CORAL to the classifier weights, leading to added efficiency when the number of classifiers is small but the number and dimensionality of target examples are very high.
- MagicVideo: Efficient Video Generation With Latent Diffusion Models517
An efficient text-to-video generation framework based on latent diffusion models, termed MagicVideo, which can generate smooth video clips that are concordant with the given text descriptions and proposes a novel VideoVAE auto-encoder for better RGB reconstruction.
- Graph-Based Global Reasoning Networks512
A new approach for reasoning globally in which a set of features are globally aggregated over the coordinate space and then projected to an interaction space where relational reasoning can be efficiently computed.
- Distilling Object Detectors With Fine-Grained Feature Imitation499
A fine-grained feature imitation method exploiting the cross-location discrepancy of feature response on the near object anchor locations reveals important information of how teacher model tends to generalize.
- 495
- Subspace Clustering by Block Diagonal Representation472
With this regularizer, the subspace clustering problem is solved by Block Diagonal Representation (BDR), which uses the block diagonal structure prior and the BDR model is nonconvex and the experiments demonstrate the effectiveness of BDR.
- 469
- MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model426
This work introduces Magic/snimate, a diffusion-based framework that aims at enhancing temporal consistency, preserving reference image faithfully, and improving animation fidelity, and outperforms the strongest baseline on the challenging TikTok dancing dataset.
- VOLO: Vision Outlooker for Visual Recognition422
A new simple and generic architecture, termed Vision Outlooker (VOLO), which implements a novel outlook attention operation that dynamically conduct the local feature aggregation mechanism in a sliding window manner across the input image, which can more efficiently encode fine-level features that are essential for high-performance visual recognition.
- Evaluation of Combined Artificial Intelligence and Radiologist Assessment to Interpret Screening Mammograms409
This diagnostic accuracy study evaluates whether artificial intelligence can overcome human mammography interpretation limits with a rigorous, unbiased evaluation of machine learning algorithms.
- 397
- Central Similarity Quantization for Efficient Image and Video Retrieval392
The Central Similarity Quantization (CSQ) is proposed, a new similarity metric with which the hash codes of similar data pairs are encouraged to approach a common center and those for dissimilar pairs to converge to different centers, to improve hash learning efficiency and retrieval accuracy.
- MetaFormer Baselines for Vision384
The capacity of MetaFormer is explored, again, by migrating the focus away from the token mixer design, and several baseline models under MetaFormer are introduced using the most basic or common mixers, and demonstrate their gratifying performance.
- PLLaVA : Parameter-free LLaVA Extension from Images to Videos for Video Dense Captioning377
This paper investigates a straight-forward, highly efficient, and resource-light approach to adapting an existing image-language pre-trained model for dense video understanding, and proposes a simple but effective pooling strategy to smooth the feature distribution along the temporal dimension and thus reduce the dominant impacts from the extreme features.
- Source Data-absent Unsupervised Domain Adaptation through Hypothesis Transfer and Labeling Transfer370
A novel approach called Source HypOthesis Transfer (SHOT), which learns the feature extraction module for the target domain by fitting the target data features to the frozen source classification module (representing classification hypothesis), and which achieves results surpassing or comparable to the state-of-the-arts.
- Semantic Object Parsing with Graph LSTM370
The Graph Long Short-Term Memory network is proposed, which is the generalization of LSTM from sequential data or multi-dimensional data to general graph-structured data.
- 343
- Shunted Self-Attention via Multi-Scale Token Aggregation339
A novel and generic strategy, termed shunted self-attention (SSA), that allows ViTs to model the attentions at hybrid scales per attention layer and out-performs the state-of-the-art Focal Transformer on Ima-geNet with only half of the model size and computation cost.
- PPDM: Parallel Point Detection and Matching for Real-Time Human-Object Interaction Detection337
A single-stage Human-Object Interaction (HOI) detection method that has outperformed all existing methods on HICO-DET dataset at 37 fps on a single Titan XP GPU is proposed, and is the first real-time HOI detection method.
- A survey on deep learning-based fine-grained object classification and semantic segmentation337
Four types of deep learning based fine-grained image classification approaches are reviewed, including the general convolutional neural networks (CNNs), part detection based, ensemble of networks based and visual attention based fine -grainedimage classification approaches.
- 334
- Overcoming Classifier Imbalance for Long-Tail Object Detection With Balanced Group Softmax318
This work provides the first systematic analysis on the underperformance of state-of-the-art models in front of long-tail distribution and proposes a novel balanced group softmax (BAGS) module for balancing the classifiers within the detection frameworks through group-wise training.
- PixelLM: Pixel Reasoning with Large Multimodal Model313
This work introduces PixelLM, an effective and efficient LMM for pixel-level reasoning and understanding, and proposes a target refinement loss to enhance the model's ability to differentiate between multiple targets, leading to substantially improved mask quality.
- Online Robust PCA via Stochastic Optimization307
An Online Robust PCA (OR-PCA) is developed that processes one sample per time instance and hence its memory cost is independent of the number of samples, significantly enhancing the computation and storage efficiency.
- Conv2Former: A Simple Transformer-Style ConvNet for Visual Recognition290
A deep look at the internal structure of self-attention is taken and a simple Transformer style convolutional neural network (ConvNet) is presented for visual recognition and a family of hierarchical ConvNets using the proposed convolutional modulation, termed Conv2Former is built.
- Hi, magic closet, tell me what to wear!290
This paper collects a large clothing What-to-Wear dataset, and thoroughly annotates the whole dataset with 7 multi-value clothing attributes and 10 occasion categories via Amazon Mechanic Turk, to learn a generalize-well model and comprehensively evaluate it.
- Video Depth Anything: Consistent Depth Estimation for Super-Long Videos277
Video Depth Anything is proposed for high-quality, consistent depth estimation in super-long videos (over several minutes) without sacrificing efficiency, and sets a new state-of-the-art in zero-shot video depth estimation.
- Rethinking Bottleneck Structure for Efficient Mobile Network Design275
This paper proposes to flip the structure and present a novel bottleneck design, called the sandglass block, that performs identity mapping and spatial transformation at higher dimensions and thus alleviates information loss and gradient confusion effectively and adds it into the search space of neural architecture search method DARTS.
- Deep neural networks and kernel regression achieve comparable accuracies for functional connectivity prediction of behavior and demographics274
This study suggests that kernel regression is as effective as DNNs for RSFC-based behavioral prediction, while incurring significantly lower computational costs, therefore, kernel regression might serve as a useful baseline algorithm for future studies.
- Understanding The Robustness in Vision Transformers268
This paper examines the role of self-attention in learning robust representations in Vision Transformers and proposes a family of fully attentional networks (FANs) that strengthen this capability by incorporating an attentional channel processing design.
- Deep subspace clustering with sparsity prior267
This paper proposes a novel subspace clustering method -- deeP subspAce clusteRing with sparsiTY prior with PARTY -- based on a new deep learning architecture that explicitly learns to progressively transform input data into nonlinear latent space and to be adaptive to the local and global subspace structure simultaneously.
- 260
- 260
- Single-Stage Multi-Person Pose Machines255
The first single-stage model, Single-stage multi-person Pose Machine (SPM), is presented to simplify the pipeline and lift the efficiency for multi- person pose estimation, and a novel Structured Pose Representation (SPR) is proposed that unifies person instance and body joint position representations.
- Attentive Contexts for Object Detection249
This work develops a novel attention to context convolution neural network (AC-CNN)-based object detection model that effectively incorporates global and local contextual information into the region-based CNN (e.g., fast R-CNN and faster R- CNN) detection framework and provides better object detection performance.
- How Far is Video Generation from World Model: A Physical Law Perspective244
This study suggests that scaling alone is insufficient for video generation models to uncover fundamental physical laws, despite its role in Sora's broader success.
- Deep Learning with S-Shaped Rectified Linear Activation Units236
A novel S-shaped rectifiedlinear activation unit (SReLU) to learn both convex and non-convex functions, imitating the multiple function forms given by the two fundamental laws, namely the Webner-Fechner law and the Stevens law, in psychophysics and neural sciences is proposed.
- Multi-fiber Networks for Video Recognition235
This paper presents the novel Multi-Fiber architecture that slices a complex neural network into an ensemble of lightweight networks or fibers that run through the network, and incorporates multiplexer modules to facilitate information flow between fibers.
- Boosting Few-Shot Learning With Adaptive Margin Loss232
An adaptive margin principle is proposed to improve the generalization ability of metric-based meta-learning approaches for few-shot learning problems by developing a class-relevant additive margin loss, where semantic similarity between each pair of classes is considered to separate samples in the feature embedding space from similar classes.
- Foreground-Aware Pyramid Reconstruction for Alignment-Free Occluded Person Re-Identification228
An alignment-free matching approach namely Foreground-aware Pyramid Reconstruction (FPR) is developed to accurately compute matching scores between occluded persons, regardless of their different scales and sizes and a occlusion-sensitive foreground probability generator is designed that focuses more on clean human body parts to robustify the similarity computation with less contamination from occlusions.
- The Devil Is in Classification: A Simple Framework for Long-Tail Instance Segmentation219
This work systematically investigates performance drop of the state-of-the-art two-stage instance segmentation model Mask R-CNN on the recent long-tail LVIS dataset, and unveils that a major cause is the inaccurate classification of object proposals.
- Robust Facial Landmark Detection via Recurrent Attentive-Refinement Networks217
A novel Recurrent Attentive-Refinement network for facial landmark detection under unconstrained conditions, suffering from challenges like facial occlusions and/or pose variations, which demonstrates superior performance in detecting challenging landmarks in comprehensive experiments and it also establishes new state-of-the-arts on the 300-W, COFW and AFLW benchmark datasets.
- 213
- Correlation Adaptive Subspace Segmentation by Trace Lasso213
The Correlation Adaptive Subspace Segmentation (CASS) method is proposed, a data correlation dependent method which simultaneously performs automatic data selection and groups correlated data together and can be regarded as a method which adaptively balances SSC and LSR.
- Predicting Alzheimer's disease progression using deep recurrent neural networks210
A recurrent neural network (RNN) model was proposed and applied to data from The Alzheimer’s Disease Prediction Of Longitudinal Evolution (TADPOLE) challenge, comprising longitudinal data of 1677 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and compared the performance of the RNN model and three baseline algorithms up to 6 years into the future.
- 209
- Deep Self-Taught Learning for Weakly Supervised Object Localization207
A deep self-taught learning approach is proposed, which makes the detector learn the object-level features reliable for acquiring tight positive samples and afterwards re-train itself based on them, which progressively improves its detection ability and localizes more informative positive samples.
- 199
- Multi-View Image Generation from a Single-View193
This paper proposes a novel image generation model termed VariGANs, which combines the merits of the variational inference and the Generative Adversarial Networks (GANs), and generates the target image in a coarse-to-fine manner instead of a single pass which suffers from severe artifacts.
- 192
- Domain Adaptation with Auxiliary Target Domain-Oriented Classifier190
A new pseudo-labeling framework called Auxiliary Target Domain-Oriented Classifier (ATDOC) is proposed, which alleviates the classifier bias by introducing an auxiliary classifier for target data only, to improve the quality of pseudo labels.
- Semantic Object Parsing with Local-Global Long Short-Term Memory188
A novel deep Local-Global Long Short-Term Memory architecture to seamlessly incorporate short-distance and long-distance spatial dependencies into the feature learning over all pixel positions and demonstrates the significant superiority of this LG-LSTM over other state-of-the-art methods.
- 187
- 186
- 185
- Flash-VStream: Memory-Based Real-Time Understanding for Long Video Streams181
This paper presents Flash-VStream, a video-language model that simulates the memory mechanism of human that is able to process extremely long video streams in real-time and respond to user queries simultaneously and achieves state-of-the-art performance in offline scenarios as well.
- Video-Based Person Re-Identification With Accumulative Motion Context181
The experimental results demonstrate that the proposed AMOC network outperforms state-of-the-arts for video-based re-identification significantly and confirm the advantage of exploiting long-range motion context forVideo-based person re-Identification, validating the motivation evidently.
- Policy Optimization with Demonstrations180
It is shown that POfD induces implicit dynamic reward shaping and brings provable benefits for policy improvement, and can be combined with policy gradient methods to produce state-of-the-art results, as demonstrated experimentally on a range of popular benchmark sparse-reward tasks.
- PoseAug: A Differentiable Pose Augmentation Framework for 3D Human Pose Estimation178
PoseAug is a new auto-augmentation framework that learns to augment the available training poses towards a greater diversity and thus improve generalization of the trained 2D-to-3D pose estimator.
- Robust Logistic Regression and Classification178
It is proved that RoLR is robust to a constant fraction of adversarial outliers, the first result on estimating logistic regression model when the covariate matrix is corrupted with any performance guarantees.
- Partial Order Pruning: For Best Speed/Accuracy Trade-Off in Neural Architecture Search176
This work proposes an algorithm that can offer better speed/accuracy trade-off of searched networks, and presents several Dongfeng (DF) networks that provide high accuracy and fast inference speed on various application GPU platforms.
- Robust LSTM-Autoencoders for Face De-Occlusion in the Wild175
A robust LSTM-Autoencoders (RLA) model is proposed to effectively restore partially occluded faces even in the wild and provides significantly larger performance gain than other de-occlusion methods in promoting recognition performance over partially-occluded faces.
- Neural Person Search Machines171
Evaluations on two benchmark datasets have demonstrated that the Neural Person Search Machines (NPSM) developed can outperform current state-of-the-arts in both mAP and top-1 evaluation protocols.
- PSGAN: Pose and Expression Robust Spatial-Aware GAN for Customizable Makeup Transfer170
Pose and expression robust Spatial-aware GAN (PSGAN) is proposed, which not only achieves state-of-the-art results even when large pose and expression differences exist but also is able to perform partial and shade-controllable makeup transfer.
- Dual-Agent GANs for Photorealistic and Identity Preserving Profile Face Synthesis169
Experimental results show that the proposed Dual-Agent Generative Adversarial Network (DA-GAN) model not only presents compelling perceptual results but also significantly outperforms state-of-the-arts on the large-scale and challenging NIST IJB-A unconstrained face recognition benchmark.
- 169
- Sa2VA: Marrying SAM2 With MLLM for Dense Grounded Understanding of Images and Videos168
Experiments show that Sa2VA achieves strong performance across multiple tasks, particularly in referring video object segmentation, highlighting its potential for complex real-world applications.
- Diffusion Probabilistic Model Made Slim167
A customized design for slim DPM is introduced, which is term as Spectral Diffusion (SD), for light-weight image synthesis, and achieves 8–18 × computational complexity reduction as compared to the latent diffusion models on a series of conditional and unconditional image generation tasks while retaining competitive image fidelity.
- 167
- IAN: The Individual Aggregation Network for Person Search165
This work proposes a novel Individual Aggregation Network (IAN) that can accurately localize persons by learning to minimize intra-person feature variations and introduces a novel center loss that can increase the intra-class compactness of feature representations.
- Efficient Hyperparameter Optimization for Deep Learning Algorithms Using Deterministic RBF Surrogates165
A new deterministic and efficient hyperparameter optimization method that employs radial basis functions as error surrogates, called HORD, which significantly outperforms the well-established Bayesian optimization methods such as GP, SMAC, and TPE.
- TS$$^{2}$$C: Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection160
This work provides a simple approach to discover tight object bounding boxes with only image-level supervision, called Tight box mining with Surrounding Segmentation Context (TS2C), developed based on two key properties of desirable bounding box: high purity and high completeness.
- 159
- Few-Shot Adaptive Faster R-CNN156
A novel few-shot adaptive Faster-RCNN framework, termed FAFRCNN, which effectively adapts to target domain with a few labeled samples, which achieves new state-of-the-art performance under both the interested few- shot domain adaptation (FDA) and unsupervised domain adaptation(UDA) setting.
- 155
- 154
- 154
- Mutual Learning to Adapt for Joint Human Parsing and Pose Estimation153
This paper presents a novel Mutual Learning to Adapt model (MuLA) for joint human parsing and pose estimation that effectively exploits mutual benefits from both tasks and simultaneously boosts their performance.
- 152
- 150
- Recovering the Unbiased Scene Graphs from the Biased Ones139
This paper shows that, due to the missing labels, SGG can be viewed as a "Learning from Positive and Unlabeled data" (PU learning) problem, where the reporting bias can be removed by recovering the unbiased probabilities from the biased ones by utilizing label frequencies.
- BuboGPT: Enabling Visual Grounding in Multi-Modal LLMs137
BozoGPT is a multi-modal LLM with visual grounding that can perform cross-modAL interaction between vision, audio and language, providing fine-grained understanding of visual objects and other given modalities and performs consistently well when provided by arbitrary modalities.
- Cycle-SUM: Cycle-Consistent Adversarial LSTM Networks for Unsupervised Video Summarization136
Experiments on two video summarization benchmark datasets validate the state-of-theart performance and superiority of the Cycle-SUM model over previous baselines.
- Learning Markov Clustering Networks for Scene Text Detection136
A novel framework named Markov Clustering Network (MCN) is proposed for fast and robust scene text detection that generates bounding boxes without using Non-Maximum Suppression, and it can be fully parallelized on GPUs.
- 136
- 134
- Attention-Aware Deep Adversarial Hashing for Cross-Modal Retrieval134
An adversarial hashing network with an attention mechanism to enhance the measurement of content similarities by selectively focusing on the informative parts of multi-modal data.
- Exact Low Tubal Rank Tensor Recovery from Gaussian Measurements130
It is shown that by solving a TNN minimization problem, the underlying tensor of size n1×n2×n3 with tubal rank r can be exactly recovered when the given number of Gaussian measurements is O(r(n1+n2−r)n3).
- PnP-DETR: Towards Efficient Visual Analysis with Transformers129
This work encapsulates the idea of reducing spatial redundancy into a novel poll and pool (PnP) sampling module, with which it is built an end-to-end PnP-DETR architecture that adaptively allocates its computation spatially to be more efficient.
- 127
- A Balanced and Uncertainty-Aware Approach for Partial Domain Adaptation126
A novel domain adaptation method BA$3$US is proposed that emphasizes uncertain samples and exploits an adaptive weighted complement entropy objective to encourage incorrect classes to have uniform and low prediction scores.
- 126
- Zigzag Learning for Weakly Supervised Object Detection125
This paper proposes a zigzag learning strategy to simultaneously discover reliable object instances and prevent the model from overfitting initial seeds, and introduces a novel masking regularization strategy over the high level convolutional feature maps to avoid over fitting initial samples.
- 122
- DINE: Domain Adaptation from Single and Multiple Black-box Predictors120
A new two-step knowledge adaptation framework called DIstill and fine-tuNE (DINE) is proposed, which distills the knowledge from the source predictor to a customized target model, then fine-tunes the distilled model to further fit the target domain.
- 120
- Video Scene Parsing with Predictive Feature Learning117
It is experimentally proved that the learned predictive features in the model are able to significantly enhance the video parsing performance by combining with the standard image parsing network.
- Understanding and Resolving Performance Degradation in Deep Graph Convolutional Networks114
A variance-controlling technique termed Node Normalization (NodeNorm), which scales each node's features using its own standard deviation, which enables deep GCNs to outperform shallow ones in cases where deep models are needed, and to achieve comparable results with shallow ones on 6 benchmark datasets.
- 112
- Prompting Depth Anything for 4K Resolution Accurate Metric Depth Estimation110
This work uses a low-cost LiDAR as the prompt to guide the Depth Anything model for accurate metric depth output, achieving up to 4K resolution, and proposes a scalable data pipeline that includes synthetic data LiDAR simulation and real data pseudo GT depth generation.
- Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoders110
Dora-Vae is presented, a novel approach that enhances VAE reconstruction through the authors' proposed sharp edge sampling strategy and a dual cross-attention mechanism that enable the VAE to focus on crucial geometric details that are typically missed by uniform sampling approaches.
- Loong: Generating Minute-level Long Videos with Autoregressive Language Models110
This work proposes Loong, a new autoregressive LLM-based video generator that can generate minute-long videos and proposes progressive short-to-long training with a loss re-weighting scheme to mitigate the loss imbalance problem for long video training.
- 108
- 106
- Learning Generalizable and Identity-Discriminative Representations for Face Anti-Spoofing105
A new generalizable face authentication CNN (GFA-CNN) model with three novelties, which introduces a simple yet effective total pairwise confusion loss for CNN training that properly balances contributions of all spoofing patterns for recognizing the spoofing faces.
- Look across Elapse: Disentangled Representation Learning and Photorealistic Cross-Age Face Synthesis for Age-Invariant Face Recognition104
This work proposes a deep Age-Invariant Model (AIM) for face recognition in the wild with three distinct novelties, and proposes a new large-scale Cross-Age Face Recognition (CAFR) benchmark dataset to facilitate existing efforts and push the frontiers of age-invariant face recognition research.
- Interpretable Structure-Evolving LSTM101
The effectiveness of structure-evolving LSTM in the application of semantic object parsing is evaluated and it is demonstrated to have an advantage over state-of-the-art L STM models on standard benchmarks.
- Revisit Knowledge Distillation: a Teacher-free Framework100
It is argued that the success of KD is not fully due to the similarity information between categories, but also to the regularization of soft targets, which is equally or even more important.
- Revisiting Temporal Modeling for CLIP-Based Image-to-Video Knowledge Transferring99
Spatial-Temporal Auxiliary Network (STAN) is presented - a simple and effective temporal modeling mechanism extending CLIP model to diverse video tasks that adopts a branch structure with decomposed spatial-temporal modules that enable multi-level CLIP features to be spatial-Temporally contextualized.
- Meta-matching as a simple framework to translate phenotypic predictive models from big to small data99
It is demonstrated that meta-matching can greatly boost the prediction of new phenotypes in small independent datasets in many scenarios, and a lower bound on the potential of meta- matching is presented.
- Efficient Meta Learning via Minibatch Proximal Update99
This paper proposes a minibatch proximal update based meta-learning approach for learning to efficient hypothesis transfer and theoretically justifies the benefit of the learnt prior hypothesis for fast adaptation to new few-shot learning tasks via minibatches proximalupdate.
- Dynamic Feature Fusion for Semantic Edge Detection96
This work proposes a novel dynamic feature fusion strategy that assigns different fusion weights for different input images and locations adaptively, and outperforms all existing well established methods and achieves new state-of-the-art methods.
- Vista-llama: Reducing Hallucination in Video Language Models via Equal Distance to Visual Tokens92
This work proposes Vista-llama, a novel framework that maintains the consistent distance between all visual tokens and any language tokens, irrespective of the generated text length, and presents a sequential visual projector that projects the current video frame into tokens of language space with the assistance of the previous frame.
- Seaweed-7B: Cost-Effective Training of Video Generation Foundation Model90
This technical report presents a cost-efficient strategy for training a video generation foundation model called Seaweed-7B trained from scratch using 665,000 H100 GPU hours and highlights the key design decisions that enhance the performance of the medium-sized diffusion model.
- Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning89
Class-wise Decorrelation (CwD) is proposed that regularizes representations of each class to scatter more uniformly, thus mimicking the model jointly trained with all classes (i.e., the oracle model), and is simple to implement and easy to plug into existing methods.
- 89
- Left-Right Comparative Recurrent Model for Stereo Matching89
A novel left- right comparative recurrent model to perform left-right consistency checking jointly with disparity estimation and a soft attention mechanism is introduced, which employs the learned error maps for better guiding the model to selectively focus on refining the unreliable regions at the next recurrent step.
- Modality-Dependent Cross-Media Retrieval89
This article proposes a modality-dependent cross-media retrieval (MDCR) model, where two couples of projections are learned for different cross- media retrieval tasks instead of one couple of projections, and achieves a new state-of-the-art performance on the Wikipedia dataset.
- Flash-Vstream: Efficient Real-Time Understanding for Long Video Streams88
This work proposes Flash-VStream, an efficient video language model capable of processing extremely long videos and responding to user queries in real time, and achieves significant reductions in inference latency.
- 85
- Expanding Small-Scale Datasets with Guided Imagination84
A Guided Imagination Framework that leverages cutting-edge generative models like DALL-E2 and Stable Diffusion to "imagine" and create informative new data from the input seed data to expand a ready-to-use small dataset by automatically creating new labeled samples.
- 84
- Beyond Object Proposals: Random Crop Pooling for Multi-Label Image Recognition84
This paper proposes an object-proposal-free framework for multi-label image recognition: random crop pooling (RCP), which performs stochastic scaling and cropping over images before feeding them to a standard convolutional neural network, which works quite well with a max-pooling operation for recognizing the complex contents of multi- label images.
- MagicEdit: High-Fidelity and Temporally Coherent Video Editing83
It is found that high-fidelity and temporally coherent video-to-video translation can be achieved by explicitly disentangling the learning of content, structure and motion signals during training.
- 3D-Aided Deep Pose-Invariant Face Recognition83
A 3D-Aided Deep Pose-Invariant Face Recognition Model (3D-PIM), which automatically recovers realistic frontal faces from arbitrary poses through a 3D face model in a novel way, and demonstrates superiority of the proposed model over state-of-the-arts.
- MagicMix: Semantic Mixing with Diffusion Models82
MagicMix is presented, a simple yet effective solution based on pre-trained text-conditioned diffusion models that does not require any spatial mask or re-training, yet is able to synthesize novel objects with high fidelity.
- Finding any Waldo with zero-shot invariant and efficient visual search81
It is shown for the first time that humans can efficiently and invariantly search for natural objects in complex scenes and a biologically-inspired zero-shot model that captures human eye movements during search is introduced.
- 80
- Adversarial Self-supervised Learning for Semi-supervised 3D Action Recognition80
Adversarial Self-Supervised Learning (ASSL) is presented, a novel framework that tightly couples SSL and the semi-supervised scheme via neighbor relation exploration and adversarial learning to improve the discrimination capability of learned representations for 3D action recognition.
- 80
- Dynamic Kernel Distillation for Efficient Pose Estimation in Videos78
A novel Dynamic Kernel Distillation model is proposed to facilitate small networks for estimating human poses in videos, thus significantly lifting the efficiency and its state-of-the-art accuracy.
- 76
- Joint Face Image Restoration and Frontalization for Recognition75
Qualitative and quantitative experiments demonstrate the superiority of MDFR over state-of-the-art methods on both face frontalization and face restoration and can generate frontalized high-quality face images by a single unified network, showing a strong capability of preserving face identity.
- Pose Partition Networks for Multi-person Pose Estimation75
This paper proposes a novel Pose Partition Network (PPN) to address the challenging multi-person pose estimation problem and implements PPN with the Hourglass architecture as the backbone network to simultaneously learn joint detector and dense regressor.
- Body Meshes as Points74
A new method that represents multiple person instances as points in the spatial-depth space where each point is associated with one body mesh, Body Meshes as Points (BMP), to simplify the pipeline and lift both efficiency and performance.
- 74
- Efficient Hyperparameter Optimization of Deep Learning Algorithms Using Deterministic RBF Surrogates72
- 71
- The Alzheimer's Disease Prediction Of Longitudinal Evolution (TADPOLE) Challenge: Results after 1 Year Follow-up70
Current prediction algorithms provide sufficient accuracy to exploit biomarkers related to clinical diagnosis and ventricle volume, for cohort refinement in clinical trials for Alzheimer's disease, however, results call into question the usage of cognitive test scores for patient selection and as a primary endpoint in clinical Trials.
- 70
- 70
- Dynamic Conditional Networks for Few-Shot Learning70
This paper proposes a novel Dynamic Conditional Convolutional Network (DCCN) to handle conditional few-shot learning, i.e, only a few training samples are available for each condition, and evaluates DCCN on four tasks which can be formulated as conditional model learning.
- Dataset Quantization69
DQ is the first method that can successfully distill large-scale datasets such as ImageNet-1k with a state-of-the-art compression ratio and can be trained with negligible or no performance drop for both vision tasks as well as language tasks.
- 68
- 67
- Traceable Evidence Enhanced Visual Grounded Reasoning: Evaluation and Methodology66
TreeBench (Traceable Evidence Evaluation Benchmark), a diagnostic benchmark built on three principles: focused visual perception of subtle targets in complex scenes, traceable evidence via bounding box evaluation, and second-order reasoning to test object interactions and spatial hierarchies beyond simple object localization are proposed.
- Clover: Towards A Unified Video-Language Alignment and Fusion Model65
Clover is introduced, a Correlated Video-Language pre-training method-towards a universal Video- language model for solving multiple video understanding tasks with neither performance nor efficiency compromise, and improves cross-modal feature alignment and fusion via a novel tri- modal alignment pre- training task.
- Global Knowledge Calibration for Fast Open-Vocabulary Segmentation64
A core idea of preserving the generalizable representation when fine-tuning on known classes is proposed, and a text diversification strategy that generates a set of synonyms for each training category is introduced, which prevents the learned representation from collapsing onto specific known category names.
- DiG: Scalable and Efficient Diffusion Models with Gated Linear Attention63
This paper aims to incorporate the sub-quadratic modeling capability of Gated Linear Attention (GLA) into the 2D diffusion backbone, and introduces Diffusion Gated Linear Attention Transformers (DiG), a simple, adoptable solution with minimal parameter overhead.
- Parallelized Autoregressive Visual Generation63
This work develops a parallel generation strategy that generates distant tokens with weak dependencies in parallel while maintaining sequential generation for strongly dependent local tokens, and can be seamlessly integrated into standard autoregressive models without modifying the architecture or tokenizer.
- MagicVideo-V2: Multi-Stage High-Aesthetic Video Generation63
This work introduces MagicVideo-V2 that integrates the text-to-image model, video motion generator, reference image embedding module and frame interpolation module into an end-to-end video generation pipeline that can generate an aesthetically pleasing, high-resolution video with remarkable fidelity and smoothness.
- 63
- Zero-Shot Learning via Attribute Regression and Class Prototype Rectification63
This paper forms ZSL as an attribute regression problem, and a new objective function which takes into consideration both the attribute regression accuracy and the class prototype discrimination is proposed, so that domain shift problem and hubness problem can be mitigated.
- Image-to-Video Generation via 3D Facial Dynamics62
This paper proposes to “imagine” a face video from a single face image according to the reconstructed 3D face dynamics, aiming to generate a realistic and identity-preserving face video, with precisely predicted pose and facial expression.
- 62
- Geometry-Guided Progressive NeRF for Generalizable and Efficient Neural Human Rendering61
To better tackle self-occlusion, a geometry-guided multi-view feature integration approach that utilizes the estimated geometry prior to integrate the incomplete information from input views and construct a complete geometry volume for the target human body is devised.
- PVRED: A Position-Velocity Recurrent Encoder-Decoder for Human Motion Prediction61
A novel Position-Velocity Recurrent Encoder-Decoder (PVRED) for human motion prediction, which makes full use of pose velocities and temporal positional information and considerably outperforms the state-of-the-art methods.
- FoveaNet: Perspective-Aware Urban Scene Parsing61
This work proposes a new FoveaNet model to fully exploit the perspective geometry of scene images and address the common failures of generic parsing models, and introduces a new dense CRFs model that takes the Perspective geometry as a prior potential.
- Reversible Recursive Instance-Level Object Segmentation61
This work proposes a novel Reversible Recursive Instance-level Object Segmentation (R2-IOS) framework that consists of a reversible proposal refinement sub-network that predicts bounding box offsets for refining the object proposal locations, and an instance-level segmentation sub- network that generates the foreground mask of the dominant object instance in each proposal.
- 60
- Multistage Object Detection With Group Recursive Learning59
A new EM-like group recursive learning approach to iteratively refine object proposals by incorporating such context of surrounding proposals and provide an optimal spatial configuration of object detections is proposed.
- Video super-resolution based on spatial-temporal recurrent residual networks59
The proposed Spatial-Temporal Recurrent Residual Network (STR-ResNet) is able to efficiently reconstruct videos with diversified contents and complex motions, which outperforms the existing video SR approaches and offers new state-of-the-art performances on benchmark datasets.
- Deep Recurrent Regression for Facial Landmark Detection59
A novel end-to-end deep architecture for face landmark detection, based on a deep convolutional and deconvolutional network followed by carefully designed recurrent network structures is proposed.
- Cascade Subspace Clustering58
This work proposes a deep learning based subspace clustering method which simultaneously learns a compact representation using a neural network and a clustering assignment by minimizing the discrepancy between pair-wise sample-centers distributions.
- MetaSelector: Meta-Learning for Recommendation with User-Level Adaptive Model Selection57
This work proposes a meta-learning framework to facilitate user-level adaptive model selection in recommender systems and achieves improvements over single model baselines and sample-level model selector in terms of AUC and LogLoss.
- Dual Adversarial Autoencoders for Clustering57
This work proposes dual AAE (Dual-AAE) which simultaneously maximizes the likelihood function and mutual information between observed examples and a subset of latent variables and derives a new reconstruction loss which can be optimized by training a pair of AEs.
- 56
- VideoWorld: Exploring Knowledge Learning from Unlabeled Videos55
This study develops and introduces the Latent Dynamics Model (LDM) as a key component of VideoWorld, an auto-regressive video generation model trained on unlabeled video data, and tests its knowledge acquisition abilities in video-based Go and robotic control tasks.
- 55
- Continual Learning via Bit-Level Information Preserving55
This work proposes a novel continual learning approach called Bit-Level Information Preserving (BLIP) that preserves the information gain on model parameters through updating the parameters at the bit level, which can be conveniently implemented with parameter quantization.
- 55
- 55
- Multimodal Learning and Reasoning for Visual Question Answering53
This work introduces a modular neural network model that learns a multimodal and multifaceted representation of the image and the question and achieves a new state-of-the-art performance on both VQA benchmark datasets.
- 53
- Multi-Prototype Networks for Unconstrained Set-based Face Recognition52
A novel Multi-Prototype Network (MPNet) model is proposed to learn multiple prototype face representations adaptively from the media sets to demonstrate superiority of the proposed model over state-of-the-arts.
- Training Group Orthogonal Neural Networks with Privileged Information52
A novel group orthogonal convolutional neural network (GoCNN) that learns untangled representations within each layer by exploiting provided privileged information and enhances representation diversity effectively is proposed.
- 51
- 51
- 51
- MagicArticulate: Make Your 3D Models Articulation-Ready50
Extensive experiments demonstrate that MagicArticulate significantly outperforms existing methods across diverse object categories, achieving high-quality articulation that enables realistic animation.
- Happiness level prediction with sequential inputs via multiple regressions50
This paper presents the solution submitted to the Emotion Recognition in the Wild (EmotiW 2016) group-level happiness intensity prediction sub-challenge, which uses a convolutional neural network to extract discriminative face features, and a recurrent neuralnetwork to selectively memorize the important features to perform the group- level happiness prediction task.
- $$\ell ^{0}$$-Sparse Subspace Clustering49
It is proved that subspace-sparse representation, a key element in subspace clustering, can be obtained by \(\ell ^{0}\)-SSC for arbitrary distinct underlying subspaces almost surely under the mild i.i.d. assumption on the data generation.
- Joint Rain Detection and Removal via Iterative Region Dependent Multi-Task Learning.49
An iterative information feedback (IIF) network that removes rain streaks and clears up the rain accumulation iteratively and progressively is proposed, which constitutes a network that is end-to-end trainable.
- 48
- Divide to Adapt: Mitigating Confirmation Bias for Domain Adaptation of Black-Box Predictors48
A new method, named BETA, is proposed to incorporate knowledge distillation and noisy label learning into one coherent framework, enabled by a new divide-to-adapt strategy, and theoretically shows that the target error of BETA is minimized by decreasing the noise ratio of the subdomains.
- PoseTriplet: Co-evolving 3D Human Pose Estimation, Imitation, and Hallucination under Self-supervision47
This paper proposes a novel self-supervised approach that allows us to explicitly generate 2D-3D pose pairs for augmenting supervision, through a self-enhancing dual-loop learning framework, and achieves encouraging results significantly outperforming the state of the art and, in some cases, even on par with results of fully-super supervised methods.
- 47
- Sharing Residual Units Through Collective Tensor Factorization To Improve Deep Neural Networks47
This work revisits the standard residual function as well as its several successful variants and proposes a unified framework based on tensor Block Term Decomposition (BTD) to explain these apparently different residual functions from the tensor decomposition view.
- 47
- Robust Face Recognition with Deep Multi-View Representation Learning47
This paper describes the proposed method targeting at the MSR Image Recognition Challenge MS-Celeb-1M, to recognize one million celebrities from their face images captured in the real world, which provides a large scale dataset crawled from the Web, which contains a large number of celebrities with many images for each subject.
- PSGAN++: Robust Detail-Preserving Makeup Transfer and Removal45
Experiments demonstrate that PSGAN++ not only achieves state-of-the-art results with fine makeup details even in cases of large pose/expression differences but also can perform partial or degree-controllable makeup transfer.
- Fine-Grained Multi-human Parsing45
A new large-scale database “Multi-human Parsing (MHP v2.0)” is presented, and a novel deep Nested Adversarial Network (NAN) model for multi-human parsing is proposed, which consistently outperforms existing state-of-the-art solutions on MHP and several other datasets.
- Low-Resolution Self-Attention for Semantic Segmentation43
The Low-Resolution Self-Attention (LRSA) mechanism to capture global context at a significantly reduced computational cost, i.e., FLOPs is introduced and the effectiveness of the approach is demonstrated by building the LRFormer, a vision transformer with an encoder-decoder structure.
- Efficient Stochastic Gradient Hard Thresholding43
It is proved that the stochastic gradient evaluation complexity of HSG-HT scales linearly with inverse of sub-optimality and its hard thresholding complexity scales logarithmically.
- Online PCA for Contaminated Data43
The online robust PCA algorithm is proposed, which is able to improve the PCs estimation upon an initial one steadily, even when faced with a constant fraction of outliers, and achieves the maximal robustness with a 50% breakdown point.
- 42
- Learning with rethinking: Recurrently improving convolutional neural networks through feedback42
By adding a feedback layer and producing the emphasis vector, the model is able to recurrently boost the performance based on previous prediction, and can be employed to boost any pre-trained models.
- 42
- ℓ ^0 ℓ 0 -Sparse Subspace Clustering.41
It is proved that subspace-sparse representation, a key element in subspace clustering, can be obtained by \(\ell ^{0}\)-SSC for arbitrary distinct underlying subspaces almost surely under the mild i.i.d. assumption on the data generation.
- TAPS3D: Text-Guided 3D Textured Shape Generation from Pseudo Supervision40
A novel framework, TAPS3D, is presented to train a text-guided 3D shape generator with pseudo captions based on rendered 2D images, which can generate 3D textured shapes from the given text without any additional optimization.
- 40
- Marginalized CNN: Learning Deep Invariant Representations40
It is proved that mCNN, the marginalized Convolutional Neural Network, can be understood as a classic CNN with a special form of regularization and thus is efficient for implementation and not restricted to the CNN module used.
- 40
- Trace Anything: Representing Any Video in 4D via Trajectory Fields39
This work proposes representing any video as a Trajectory Field: a dense mapping that assigns a continuous 3D trajectory function of time to each pixel in every frame, and introduces Trace Anything, a neural network that predicts the entire trajectory field in a single feed-forward pass.
- 39
- 39
- 39
- 39
- DREAM-Talk: Diffusion-based Realistic Emotional Audio-driven Method for Single Image Talking Face Generation38
DREAM-Talk is introduced, a two-stage diffusion-based audio-driven framework, tailored for generating diverse expressions and accurate lip-sync concurrently, and outperforms state-of-the-art methods in terms of expressiveness, lip-sync accuracy and perceptual quality.
- ST-HOI: A Spatial-Temporal Baseline for Human-Object Interaction Detection in Videos38
This paper shows that a naive temporal-aware variant of a common action detection baseline does not work on video-based HOIs due to a feature-inconsistency issue, and proposes a simple yet effective architecture utilizing temporal information such as human and object trajectories, correctly-localized visual features, and spatial-temporal masking pose features.
- Cross-Domain Human Parsing via Adversarial Feature and Label Adaptation38
The proposed model explicitly learns a feature compensation network, which is specialized for mitigating the cross-domain differences, and introduces a structured label adversarial network to guide the parsing results of the target domain to follow the high-orders of the structured labels shared across domains.
- 37
- Residual Distillation: Towards Portable Deep Neural Networks without Shortcuts37
This work proposes a novel joint-training framework to train plain CNN by leveraging the gradients of the ResNet counterpart, and demonstrates that the plainCNN network without shortcuts can achieve the same level of accuracy as that of the ResNet baseline while achieving about 1 .
- Recognizing Profile Faces by Imagining Frontal View37
Qualitative and quantitative experiments on both controlled and in-the-wild benchmark datasets demonstrate the superiority of the proposed Pose-Invariant Model (PIM) for face recognition in the wild, with three distinct novelties.
- Adaptive ROI generation for video object segmentation using reinforcement learning36
A novel approach which utilizes reinforcement learning to select optimal adaptation areas for each frame, based on the historical segmentation information is proposed, which improves the state-of-the-art of the mean region similarity on DAVIS 2016 dataset to 87.1%.
- GETAvatar: Generative Textured Meshes for Animatable Human Avatars35
The proposed GETAvatar is a Generative model that directly generates Explicit Textured 3D meshes for animatable human Avatar, with photorealistic appearance and fine geometric details, and achieves state-of-the-art performance on 3D-aware human generation both in appearance and geometry quality.
- The Scalability of Simplicity: Empirical Analysis of Vision-Language Learning with a Single Transformer34
Sail is introduced, a single transformer unified multimodal large language model (MLLM) that integrates raw pixel encoding and language decoding within a singular architecture that eliminates the need for a separate vision encoder, presenting a more minimalist architecture design.
- Effective Training Strategies for Deep Graph Neural Networks.34
The proposed NodeNorm regularizes deep GCNs by discouraging feature-wise correlation of hidden embeddings and increasing model smoothness with respect to input node features, and thus effectively reduces overfitting, enabling deep GNNs to compete with and even outperform shallow ones.
- 34
- Seeing Human Weight from a Single RGB-D Image34
This article proposes a novel weight estimator based on a single RGB-D image, which utilizes the visual color cues and depth information and considers gender as another important factor for human weight estimation.
- 33
- Generalizing Few-Shot NAS with Gradient Matching33
A gradient matching score (GM) that leverages gradient information at the shared weight for making informed splitting decisions is proposed that significantly outperforms its Few-Shot counterparts while surpassing previous comparable methods in terms of the accuracy of derived architectures.
- 33
- Bridging Continuous and Discrete Tokens for Autoregressive Visual Generation32
This work introduces a dimension-wise quantization strategy that independently discretizes each feature dimension, paired with a lightweight autoregressive prediction mechanism that efficiently model the resulting large token space.
- VLAB: Enhancing Video Language Pretraining by Feature Adapting and Blending32
A novel video-text pre-training method dubbed VLAB, which transfers CLIP representations to video pre-training tasks and develops unified video multimodal models for a wide range of video-text tasks and introduces a new video adapter module to address CLIP's deficiency in modeling temporal information and extend the model's capability to encompass both contrastive and generative tasks.
- Test-Agnostic Long-Tailed Recognition by Test-Time Aggregating Diverse Experts with Self-Supervision32
- Joint 3D Face Reconstruction and Dense Face Alignment from A Single Image with 2D-Assisted Self-Supervised Learning.32
A novel 2D-assisted self-supervised learning (2DASL) method that can effectively use "in-the-wild" 2D face images with noisy landmark information to substantially improve 3D face model learning and significantly relieves demands on the conventional paired 2D to 3D annotations.
- Transferable Meta Learning Across Domains32
This work introduces the cross-domain meta learning framework and proposes a new trans-ferable meta learning (TML) algorithm, which performs meta task adaptation jointly with meta model learning, which effectively nar-rows divergence between source and target tasks and enables transferring source meta-knowledge to solve target tasks.
- Multi-Path Feedback Recurrent Neural Networks for Scene Parsing32
This paper considers the scene parsing problem and proposes a novel Multi-Path Feedback recurrent neural network (MPF-RNN), which can enhance the capability of RNNs in modeling long-range context information at multiple levels and better distinguish pixels that are easy to confuse.
- 32
- 31
- 30
- 30
- 29
- 29
- 29
- ML-LocNet: Improving Object Localization with Multi-view Learning Network29
A Multi-view Learning Localization Network (ML-LocNet) is proposed by incorporating multi-view learning into a two-phase WSOL model, which effectively avoids early fitting and can be easily combined with existing WSOL models to further improve the localization accuracy.
- PV3D: A 3D Generative Model for Portrait Video Generation28
Pv3D is proposed, the first generative framework that can synthesize multi-view consistent portrait videos with high-quality appearance and geometry and is able to support many downstream applications such as animating static portraits and view-consistent video motion editing.
- 28
- Deep multi-person kinship matching and recognition for family photos28
The results show that the proposed Deep Kinship Matching and Recognition framework can gain superior performance on Group-Face dataset, TSKinFace dataset and FIW dataset over state-of-the-arts datasets.
- Deep Attribute-preserving Metric Learning for Natural Language Object Retrieval28
A deep Attribute-Preserving Metric learning (AP-Metric) framework that jointly generates unique query-sensitive region proposals and conducts novel cross-modal feature learning that explicitly pursues consistency over semantic attribute abstraction within both domains for deep metric learning is proposed.
- Multi-human Parsing with a Graph-based Generative Adversarial Model27
A novel multi- human parsing model named MH-Parser is introduced, which uses a graph-based generative adversarial model to address the challenges of close-person interaction and occlusion in multi-human parsing.
- 27
- 27
- Transferable Semi-Supervised Semantic Segmentation27
A novel transferable semi-supervised semantic segmentation model that can transfer the learned segmentation knowledge from a few strong categories with pixel- level annotations to unseen weak categories with only image-level annotations is proposed, significantly broadening the applicable territory of deep segmentation models.
- Multi-Human Parsing Machines27
The proposed Multi-Human Parsing Machines (MHPM) system, which contains an MHP Montage model and a MHP Solver, to address both challenges in multi-human parsing, can achieve state-of-the-art performance on the multi- human parsing benchmark and the person individualization benchmark.
- Visual Relationship Detection With Visual-Linguistic Knowledge From Multimodal Representations26
A novel approach named Relational Visual-Linguistic Bidirectional Encoder Representations from Transformers (RVL-BERT), which performs relational reasoning with both visual and language commonsense knowledge learned via self-supervised pre-training with multimodal representations.
- 26
- 26
- 26
- Integrated Face Analytics Networks through Cross-Dataset Hybrid Training26
An integrated Face Analytics Network (iFAN) is proposed, which is able to perform multiple tasks jointly for face analytics with a novel carefully designed network architecture to fully facilitate the informative interaction among different tasks.
- 26
- Pixel-SAIL: Single Transformer For Pixel-Grounded Understanding25
The Pixel-SAIL, a single transformer for pixel-wise MLLM tasks, is presented and a novel visual prompt injection strategy to enable the single transformer to understand visual prompt inputs and benefit from the early fusion of visual prompt embeddings and vision tokens is proposed.
- 25
- 25
- MagicAvatar: Multimodal Avatar Generation and Animation24
The flexibility of MagicAvatar is demonstrated through various applications, including text-guided and video-guided avatar generation, as well as multimodal avatar animation, which enables the animation of the provided human identity according to the specific motion derived from the first stage.
- Don't ask me what i'm like, just watch and listen24
It is concluded that predicting personality scores through predicting answers first is better than directly predicting scores based on audio/visual features (as studied in state-of-the art methods).
- Harnessing Diffusion Models for Visual Perception with Meta Prompts23
This paper proposes a simple yet effective scheme to harness a diffusion model for visual perception tasks, and designs a recurrent refinement training strategy that fully leverages the property of diffusion models, thereby yielding stronger visual features.
- 23
- Nonconvex Sparse Spectral Clustering by Alternating Direction Method of Multipliers and Its Convergence Analysis23
This work proposes an efficient Alternating Direction Method of Multipliers (ADMM) to solve the nonconvex SSC and provides the convergence guarantee, and proves that the sequences generated by ADMM always exist a limit point and any limit point is a stationary point.
- 23
- Collaborative Layer-Wise Discriminative Learning in Deep Neural Networks23
It is argued that if a training sample has already been correctly classified at a specific layer with high confidence, it is unnecessary to enforce rest layers to classify this sample correctly and a better strategy is to encourage those layers to focus on other samples.
- Segmentation over Detection by Coupled Global and Local Sparse Representations23
Extensive experiments on the Pascal VOC object segmentation datasets, VOC2007 and VOC2010, show that the proposed algorithm achieves competitive results with the state-of-the-art learning based algorithms, and is superior over other detection basedobject segmentation algorithms.
- Towards Adversarially Robust Deep Image Denoising22
It is found that existing DIDs are vulnerable to the adversarial noise generated by OBSATK, and an adversarial training strategy that jointly trains DIDs with adversarial and non-adversarial noisy data can recover high-quality clean images from real-world noise even without training on real noisy data.
- 22
- 22
- Learning to Localize Objects with Noisy Labeled Instances22
This paper addresses Weakly Supervised Object Localization (WSOL) with only image-level supervision by model the missing object locations as latent variables, and contributing a novel self-directed optimization strategy to infer them, which is robust to noisy initialized object locations.
- 22
- 22
- 22
- 22
- Learning Detection with Diverse Proposals22
A trainable DPP layer, allowing for Learning Detection with Diverse Proposals (LDDP), considers both label-level contextual information and spatial layout relationships between proposals without increasing the number of parameters of the network, and thus improves location and category specifications of final detected bounding boxes substantially during both training and inference schemes.
- 22
- AvatarStudio: High-Fidelity and Animatable 3D Avatar Creation from Text21
Av AvatarStudio is proposed, a generative model that yields explicit textured 3D meshes for animatable human avatars and incorporates articulation modeling into the explicit mesh representation to support high-resolution rendering and avatar animation.
- New Insight into Hybrid Stochastic Gradient Descent: Beyond With-Replacement Sampling and Convexity21
This paper affirmatively shows that under WoRS and for both convex and non-convex problems, it is still possible for HSGD (with constant step-size) to match full gradient descent in rate of convergence, while maintaining comparable sample-size-independent incremental first-order oracle complexity to stochastic gradient descent.
- 20
- MaskDiffusion: Boosting Text-to-Image Consistency with Conditional Mask20
This work identifies that a crucial factor leading to the erroneous generation of objects and their attributes is the inadequate cross-modality relation learning between the prompt and the generated images, and advances the cross-modality relation learning between the prompt and the generated images to better align the prompt and image content.
- 20
- Towards Real World Human Parsing: Multiple-Human Parsing in the Wild.20
A novel Multiple-Human Parser (MH-Parser) is proposed, which considers both the global context and local cues for each person in the parsing process and is demonstrated to outperform the naive "detect-and-parse" approach by a large margin.
- 20
- Accelerated Stochastic Mirror Descent Algorithms For Composite Non-strongly Convex Optimization20
The deterministic accelerated proximal gradient methods of Paul Tseng are extended to randomized versions for solving the problem without the strongly convex assumption and the algorithms achieve the optimal convergence rate of O(\nicefrac{1}{k^2})$.
- Learning to rank tags20
A novel semi-supervised learning framework to rank image tags is presented, which learns a ranking projection with theoretic guarantee from visual words distribution to the relevant tags distribution, and then uses it for ranking new image tags.
- MagicProp: Diffusion-based Video Editing via Motion-aware Appearance Propagation19
Overall, MagicProp combines the flexibility of image-editing techniques with the superior temporal consistency of autoregressive modeling, enabling flexible editing of object types and aesthetic styles in arbitrary regions of input videos while maintaining good temporal consistency across frames.
- 19
- ORDNet: Capturing Omni-Range Dependencies for Scene Parsing19
The proposed ORDNet outperforms previous state-of-the-art methods on three scene parsing benchmarks including PASCAL Context, COCO Stuff and ADE20K, demonstrating the superiority of capturing omni-range dependencies in deep models for scene parsing task.
- Temporal Spiking Recurrent Neural Network for Action Recognition19
The proposed TSRNN employs a novel spiking architecture which utilizes the local discriminative features from high-confidence reliable frames as spiking signals which can guide RNN to correct its long-term memory across multiple frames from contamination caused by noisy frames with distracting factors.
- 19
- 18
- Temporal Perceiving Video-Language Pre-training18
Comprehensive experimental results show that this method significantly improves the state-of-the-art performance on various benchmarks, covering text-to-video retrieval, video question answering, video captioning, temporal action localization and temporal moment retrieval.
- 18
- 18
- Dual Low-Rank Pursuit: Learning Salient Features for Saliency Detection18
This paper considers a specific saliency detection problem of predicting human eye fixations when they freely view natural images, and proposes a novel dual low-rank pursuit (DLRP) method that learns saliency-aware feature transformations by utilizing available supervision information and constructs discriminative bases for effectively detecting human fixation points under the popular low- rank and sparsity-pursuit framework.
- Multi-class learning from class proportions18
The overall solution can not only directly predict the labels of the training and new samples, but also gracefully utilize the test samples to further boost the classification performance in a manner of semi-supervised learning.
- 17
- 17
- Hyperparameter Optimization of Deep Neural Networks Using Non-Probabilistic RBF Surrogate Model.17
This work proposes to employ radial basis function as the surrogate of the error functions for optimizing both continuous and integer hyperparameters, and proposes a non-probabilistic algorithm, called Hyperparameter Optimization using RBF and DYCORS, which significantly outperforms the well-established Bayesian optimization methods.
- ManiCLIP: Multi-attribute Face Manipulation from Text16
A novel multi-attribute face manipulation method based on textual descriptions that generates natural manipulated faces with minimal text-irrelevant attribute editing and encourages the model to edit the latent code of each attribute separately via an entropy constraint.
- CMAE-V: Contrastive Masked Autoencoders for Video Action Recognition16
This work shows that CMAE also trivially generalizes well on video action recognition without modifying the architecture and the loss criterion, by directly replacing the original pixel shift with the temporal shift.
- Generative Attention Model with Adversarial Self-learning for Visual Question Answering16
This paper proposes a novel generative attention model obtained by adversarial self-learning that produces more diverse visual attention maps and it is able to generalize the attention better to new questions.
- COSA: Concatenated Sample Pretrained Vision-Language Foundation Model15
COSA, a COncatenated SAmple pretrained vision-language foundation model, jointly models visual contents and event-level temporal cues using only image-text corpora and achieves state-of-the-art results on various competitive benchmarks.
- 14
- 14
- 14
- 14
- PANDA: Prototypical Unsupervised Domain Adaptation.14
This work attempts to calibrate the noisy pseudo labels with prototypes, and develops an unified framework termed PrototypicAl uNsupervised Domain Adaptation (PANDA) for UDA, which achieves state-of-the-art or competitive results on multiple UDA benchmarks including both object recognition and semantic segmentation tasks.
- 14
- 14
- 14
- 14
- 14
- Auto-generation of professional background music for home-made videos14
A joint probabilistic framework is proposed to model prior knowledge from two aspects, namely, the correlation between video and audio tracks, as well as the transition mode horn speech and music components for background music auto-generation.
- VideoWorld 2: Learning Transferable Knowledge from Real-world Videos13
VideoWorld 2 is presented, which extends VideoWorld and offers the first investigation into learning transferable knowledge directly from raw real-world videos, with all code, data, and models to be open-sourced for further research.
- Magic-Boost: Boost 3D Generation with Multi-View Conditioned Diffusion13
This work focuses on the fine-tuning of pre-trained 2D diffusion models to harness their capacity for producing multi-view images, which are then lifted into accurate 3D models via methods like fast-NeRFs or large reconstruction models.
- AdjointDPM: Adjoint Sensitivity Method for Gradient Backpropagation of Diffusion Probabilistic Models13
A novel method AdjointDPM is proposed, which first generates new samples from diffusion models by solving the corresponding probability-flow ODEs and uses the adjoint sensitivity method to backpropagate the gradients of the loss to the models' parameters by solving another augmented ODE.
- Reachability-Aware Laplacian Representation in Reinforcement Learning13
A Reachability-Aware Laplacian Representation (RA-LapRep) is introduced, which can better capture the inter-state reachability as compared to LapRep, through both theoretical explanations and experimental results.
- 13
- Neural Epitome Search for Architecture-Agnostic Network Compression13
This work presents a novel auto-sampling method that is applicable to both 1D and 2D CNNs with significant performance improvement over WSNet and outperforms some neural architecture search (NAS) based methods such asAMC and MNasNet.
- Accelerated Randomized Mirror Descent Algorithms for Composite Non-strongly Convex Optimization13
This work considers the problem of minimizing the sum of an average function of a large number of smooth convex components and a general, possibly non-differentiable, convex function and proposes an accelerated randomized mirror descent method for solving this problem without the strongly convex assumption.
- Prototype Reminding for Continual Learning.13
This work proposes a novel method, Prototype Reminding, that efficiently embeds and recalls previously learnt knowledge to tackle catastrophic forgetting issue and suggests a promising solution for few-shot continual learning.
- 13
- 13
- 13
- 12
- Co-saliency Detection with Graph Matching12
A novel graph-matching based model is proposed to integrate the visual appearance, saliency coherence, and spatial structural continuity for detecting co-saliency collaboratively and outperforms state-of-the-art baselines significantly.
- Ensemble Robustness of Deep Learning Algorithms12
Through the lens of ensemble robustness, this work reveals that a stochastic learning algorithm can generalize well as long as its sensitiveness to adversarial perturbation is bounded in average, or equivalently, the performance variance of the algorithm is small.
- Collaborative Linear Coding for Robust Image Classification12
This paper proposes a novel image representation learning method, i.e., collaborative linear coding (CLC), to alleviate the negative influence of noisy features in classifying images, and investigates the special structure of the problems and proposes an efficient alternative optimization algorithm.
- AutoSpace: Neural Architecture Search with Less Human Interference11
A novel differentiable evolutionary framework named AutoSpace is proposed, which evolves the search space to an optimal one with following novel techniques: a differentiable fitness scoring function to efficiently evaluate the performance of cells and a reference architecture to speedup the evolution procedure and avoid falling into sub-optimal solutions.
- 11
- 11
- Better Guider Predicts Future Better: Difference Guided Generative Adversarial Networks11
A novel GAN based on inter-frame difference to circumvent the difficulties of generating unseen future frames from previous obeservations and achieves state-of-the-art performance for future video prediction on UCF-101, KITTI.
- 11
- Photo to Family Tree: Deep Kinship Understanding for Nuclear Family Photos11
This paper extends kinship recognition from two persons to a nuclear family consisting of multiple persons and proposes a novel Deep Kinship Recognition (DKR) framework, which gains superior performance on both Group-Face dataset and TSKinFace dataset, compared with state-of-the-arts.
- Subspace Learning by $$\ell ^{0}$$ ℓ 0 -Induced Sparsity11
In contrast to the required assumptions, such as independence or disjointness, on subspaces for most existing sparse subspace clustering methods, it is proved that ℓ0\documentclass[12pt]{minimal} \usepackage{amsmath} á¬wasysym} £1,000,000 \use package{amsfonts} ï¬amssymb ¬1,500,000
- 11
- 10
- Hierarchical Memory for Long Video QA10
This paper describes the champion solution to the LOVEU Challenge @ CVPR'24, Track 1 (Long Video VQA), adopting a hierarchical memory mechanism named STAR Memory, proposed in Flash-VStream, that is capable of processing long videos with limited GPU memory (VRAM).
- Video Recognition in Portrait Mode10
This paper has developed the first dataset dedicated to portrait mode video recognition, namely PortraitMode-400, comprising 400 fine-grained categories, and rigorous quality assurance was implemented to ensure the accuracy of human annotations.
- Towards Accurate Guided Diffusion Sampling through Symplectic Adjoint Method10
Symplectic Adjoint Guidance (SAG) is proposed, which calculates the gradient guidance in two inner stages and generates images with higher qualities compared to the baselines in both guided image and video generation tasks.
- 10
- 10
- 10
- Stochastic Primal-Dual Proximal ExtraGradient descent for compositely regularized optimization10
A stochastic variant of extra-gradient type methods, namely Stochastic Primal-Dual Proximal ExtraGradient descent (SPDPEG) is proposed, and its convergence property for both convex and strongly convex objectives is analyzed.
- Online Robust Low-Rank Tensor Learning10
This paper proposes an Online Robust Lowrank Tensor Modeling (ORLTM) approach, which dynamically explores the high-order correlations across all tensor modes for low-rank structure modeling and introduces a new dictionary learning component to analyze mixture data from multiple subspaces.
- 10
- Purposive hidden-object-game10
This work designs a so-called purposive hidden-object-game (P-HOG), which seamlessly embeds object localization into the enjoyable HOG gaming process, and is effective for collecting massive object locations with satisfying accuracy.
- PSGAN: Pose-Robust Spatial-Aware GAN for Customizable Makeup Transfer.9
A novel Pose-robust Spatial-aware GAN (PSGAN) for transferring the makeup style from a reference image to a source image that achieves state-of-the-art results on the existing datasets and is able to perform the customizable part-by-part, shade controllable and pose- Robust makeup transfer.
- 9
- 9
- 9
- 8
- 8
- Slim Scissors: Segmenting Thin Object from Synthetic Background8
Slim Scissors is developed, which enables quick extraction of elon-gated thin parts by simply brushing some coarse scribbles and is model-agnostic and seamlessly applicable to existing state-of-the-art interactive segmentation models.
- 8
- 8
- 8
- 8
- 8
- 8
- 8
- Annotation modification for fine-grained visual recognition8
This work proposes a novel approach from a new viewpoint to solve the fine-grained recognition problem, namely annotation modification, and obtains some most confusing classes of each image through an easy-to-evaluate classifier, and modifications of each annotation using the active set of annotations.
- Distilling Reverse-Mode Automatic Differentiation (DrMAD) for Optimizing Hyperparameters of Deep Neural Networks8
This work proposes to distill the knowledge of the forward pass into an shortcut path, through which the training trajectory is approximately reverse, making it feasible, for the first time, to automatically tune hundreds of thousands of hyperparameters of deep neural networks in practice.
- 8
- High Quality Human Image Animation using Regional Supervision and Motion Blur Condition7
This work first leverage regional supervision for detailed regions to enhance face and hand faithfulness and model the motion blur explicitly to further improve the appearance quality, and explores novel training strategies for high-resolution human animation to improve the overall fidelity.
- Delving Deeper into Data Scaling in Masked Image Modeling7
The study reveals that MIM can be viewed as an effective method to improve the model capacity when the scale of the training data is relatively small and MIM pre- training is data-agnostic under most scenarios, which means that the strategy of sampling pre-training data is non-critical.
- 7
- Deep Learning for Multimedia7
This panel aims to invite prestigious multimedia researchers and active deep learning practitioners to discuss the positioning of deep learning research now and in the future and presents their opinions on the following five questions.
- Egocentric Spatial Memory7
This work introduces an integrated deep neural network architecture for modeling egocentric spatial memory (ESM) and demonstrates its accurate and robust global mapping capacities in 3D virtual mazes and realistic indoor environments by comparing with several competitive baselines.
- 7
- 7
- Empowering Visual Creativity: A Vision-Language Assistant to Image Editing Recommendations6
This paper introduces Creativity-Vision Language Assistant(Creativity-VLA), a multimodal framework designed specifically for edit-instruction generation that aims to automatically generate diverse creative editing instructions from an input image and a simple prompt representing the users' under-specified editing purpose.
- DOAD: Decoupled One Stage Action Detection Network6
This work presents a decoupled onestage network dubbed DOAD, to mitigate above issues and improve the efficiency for spatio-temporal action detection and shows that the method is competitive with two-stage state-of-theart methods with significant efficiency improvement.
- 6
- Deep spatio-frequency saliency detection6
A Spatio-Frequency Network (SFNet) is proposed that exploits spatio-frequency clues to effectively enlarge the receptive fields of CNN layers and more importantly, strengthen their spatial discrimination for better saliency detection.
- 6
- A Live Face Swapper6
A deep learning-based face detector which is able to accurately detect faces in the wild and a face swapping framework which can smoothly replace the face appearance of a user in real time are proposed.
- Sense Beyond Expressions6
A dataset of personal images with comprehensively annotated cuteness scores and facial attributes is constructed to investigate this high-level concept in depth and a novel Continuous Latent Support Vector Machine method is developed to predict the cuteness score of one person given only his image.
- 6
- 5
- Central Similarity Hashing via Hadamard matrix.5
A new hash center network (HCN) that learns hashing functions by optimizing the central similarity w.r.t.\ these hash centers is devised and can be applied for both image and video hashing.
- 5
- 5
- 5
- Denser Trajectories of Anchor Points for Action Recognition5
The Denser Trajectories of Anchor Points scheme is introduced that improves state-of-the-art dense trajectories by paying attention to the selection of feature points for trajectory tracking.
- Robust Visual Object Tracking with Top-down Reasoning5
This work proposes to exploit the category-specific semantics to boost visual object tracking, and develops a new visual tracking model that augments the appearance based tracker with a top-down reasoning component that substantially enhances the robustness of the tracker and provides state-of-the-art performance.
- 5
- Auto-Grouped Sparse Representation for Visual Analysis5
This work investigates how to automatically uncover the underlying group structure of a feature vector such that each group characterizes certain object-specific patterns, e.g., visual pattern or motion trajectories from one object by proposing a novel auto-grouped sparse representation (ASR) method.
- 5
- ChatAnything: Facetime Chat with LLM-Enhanced Personas4
This technical report targets generating anthropomorphized personas for LLM-based characters in an online manner, including visual appearance, personality and tones, with only text descriptions, by leveraging the in-context learning capability of LLMs for personality generation by carefully designing a set of system prompts.
- Associating Spatially-Consistent Grouping with Text-supervised Semantic Segmentation4
This work investigates performing semantic segmentation solely through the training on image-sentence pairs by adapting a text-supervised model from image-level to region-level recognition with two core designs and adopts a contextually aware masking strategy to enable simultaneous recognition of all grouped regions.
- Adversarial images for the primate brain4
This work designed adversarial images to fool primate vision and modified images to match their model-predicted neuronal responses to a target category, such as monkey faces, showing that a model of neuronal activity can selectively direct primate visual behavior.
- 4
- Generalized Majorization-Minimization for Non-Convex Optimization4
This work gives the first non-asymptotic convergence analysis for MM-alike algorithms in general non-convex optimization in general majorization-Minimization problems.
- Towards a universal detector by mining concepts with small semantic gaps4
This work presents the first research attempting to substantiate the semantic gap measuring of a large amount of concepts and leverage visually learnable concepts to predicate those with no training images available.
- Image segmentation with patch-pair density priors4
A simple yet efficient multi-task joint sparse representation model is presented to augment the patch-pair similarities by harnessing the newly discovered patch- Pair density priors, which clearly demonstrate the superiority of the proposed solution over several state-of-the-art algorithms.
- 3
- 3
- 3
- EPIM: Efficient Processing-In-Memory Accelerators based on Epitome3
The Epitome is introduced, a lightweight neural operator offering convolution-like functionality, to craft memory-efficient CNN operators for PIM accelerators (EPIM), and a PIM-aware layer-wise design method to enhance their hardware efficiency is introduced.
- Learning Box Regression and Mask Segmentation Under Long-Tailed Distribution with Gradient Transfusing3
This work develops a cross class gradient transfusing (CRAT) approach to transfer the abundant training signal from head classes to help the training of sample-scarce tail classes and improves the state-of-the-art long-tailed object detection and instance segmentation models.
- 3
- 3
- 3
- 3
- 3
- 3
- 3
- Lift-the-Flap: Context Reasoning Using Object-Centered Graphs.3
The model makes reasonable inferences compared to humans, and significantly outperforms all the null models, and demonstrates the usefulness of the object-centered graph network model in context-aware object recognition and target priming in visual search.
- 3
- 4th PVUW MeViS 3rd Place Report: Sa2VA2
This report shows that with a simple modification to the test time inference method on stronger MLLMs, it can lead to stronger results on MeVIS, and adopts the recent method Sa2VA, a unified model for dense grounded understanding of both images and videos.
- 2
- 2
- Tyger: Task-Type-Generic Active Learning for Molecular Property Prediction2
A Task-type-generic active learning framework (termed Tyger) that is able to handle different types of learning tasks in a unified manner that consistently achieves high AL performance on molecular property prediction, outperforming baselines by a large margin is proposed.
- 2
- SODAR: Exploring Locally Aggregated Learning of Mask Representations for Instance Segmentation2
A novel learning-based aggregation method that improves upon SOLO by leveraging the rich neighboring information while maintaining the architectural efficiency, and is named SODAR, which significantly improves the instance segmentation performance.
- Combating Domain Shift with Self-Taught Labeling.2
Self-Taught Labeling (SeTL) is proposed, a new regularization approach that finds an auxiliary target-specific classifier for unlabeled data and significantly outperforms existing domain alignment techniques on a large variety of domain adaptation benchmarks.
- 2
- Deep Model Compression via Filter Auto-sampling.2
This work presents a novel auto-sampling method applicable to both 1D and 2D CNNs with significant performance improvement over WSNet and outperforms some neural architecture search (NAS) based methods such as AMC and MNasNet.
- 2
- What am I searching for2
This work developed a model (InferNet) whose goal was to infer what the target was and successfully identifies the subject's goal and outperforms all the competitive null models, even without any object-specific training on the inference task.
- 2
- Deep Reinforcement Learning for Accelerating the Convergence Rate2
This work proposes a principled deep reinforcement learning approach that is able to accelerate the convergence rate of general deep neural networks (DNNs) and uses deep RL to learn how to optimize a large-sized DNN.
- 1
- 1
- 1
- 1
- Class Prototype-based Cleaner for Label Noise Learning1
Unlike previous works treating all the classes equally, CPC takes advantage of loss distribution modeling and intra-class consistency regularization in feature space simultaneously and thus can better distinguish clean and noise labels.
- 1
- 1
- RVL-BERT: Visual Relationship Detection with Visual-Linguistic Knowledge from Pre-trained Representations1
A novel approach named Relational Visual-Linguistic Bidirectional Encoder Representations from Transformers (RVL-BERT), which performs relational reasoning with both visual and language commonsense knowledge learned via self-supervised pre-training with multimodal representations and decouples object detection from visual relationship recognition.
- 1
- 1
- Learning $\ell^{0}$-Graph for Data Clustering.1
A novel $\ell^{0}$-graph is proposed that employs $\ell^0$-norm to encourage the sparsity of the constructed graph, and a proximal method to solve the associated optimization problem with the proved guarantee of convergence is developed.
- 1
- Purposive hidden-object game (P-HOG) towards imperceptible human computation1
The comprehensive experiments shows that P-HOG appeals to general players and can easily perform quality control, and hence effective for collecting massive object locations with high accuracy, and may benefit general learning-based algorithms for multimedia tasks.
- –
- –
- –
- –
- –
- –
- –
- LV-BERT: Exploiting Layer Variety for BERT–
This paper aims to improve pre-trained models by exploiting layer variety from two aspects: the layer type set and the layer order, and introduces convolution into the layertype set, which is experimentally found beneficial to pre- trained models.
- –
- –
- –
- RAIN: Robust and Accurate Classification Networks with Randomization and Enhancement–
This paper proposes a novel framework to improve robustness and meanwhile retain the accuracy of given classification CNN models, termed as RAIN, which consists of two conjugate modules: structured randomization (SRd) and detail generation (DG).
- –
- –
- –
- –
- –
- –
- Central Similarity Hashing for Efficient Image and Video Retrieval–
The Central Similarity Hashing (CSH) is proposed, a new similarity metric with which the hash codes for similar data pairs are encouraged to approach a common center and those for dissimilar pairs to converge to different centers to improve hash learning efficiency and retrieval accuracy.
- Interpreting Deep Classification Models With Bayesian Inference–
A novel approach to interpret a well-trained classification model through systematically investigating effects of its hidden units on prediction making, model such a process of unit selection as an Indian Buffet Process, and derive a simplified objective function via the MAP asymptotic technique.
- Egocentric Spatial Memory Network–
Comprehensive analysis of the first deep neural network architecture for modeling Egocentric Spatial Memory suggests the essential role of individual modules in the proposed architecture and demonstrates efficiency of communications among these modules.
- WSNet: Learning Compact and Efficient Networks with Weight Sampling–
It is demonstrated that such a novel weight sampling approach (and induced WSNet) promotes both weights and computation sharing favorably and can more efficiently learn much smaller networks with competitive performance compared to baseline networks with equal numbers of convolution filters.
- –
- –
- –
- –
- –
- –
- Perception Preserving Projections–
Perception Preserving Projections (PPP) is proposed, which incorporates domain-specific feature extractor into the standard PCA formulation for the projection learning procedure, and enables PPP to make more sensible projections for feature based perception systems while retaining the simplicity and unsupervised manner of PCA.
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.