Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Bo ZhangSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works221 from public data
- CenterNet: Keypoint Triplets for Object Detection3,587
This paper presents an efficient solution that explores the visual patterns within individual cropped regions with minimal costs, and builds the framework upon a representative one-stage keypoint-based detector named CornerNet, which improves both precision and recall.
- CenterNet: Object Detection with Keypoint Triplets3,563
This paper presents an efficient solution that explores the visual patterns within individual cropped regions with minimal costs, and builds the framework upon a representative one-stage keypoint-based detector named CornerNet, which improves both precision and recall.
- Accurate medium-range global weather forecasting with 3D neural networks2,041
It is shown that three-dimensional deep networks equipped with Earth-specific priors are effective at dealing with complex patterns in weather data, and that a hierarchical temporal aggregation strategy reduces accumulation errors in medium-range forecasting.
- 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering1,670
This work proposes 4D Gaussian Splatting (4D-GS) as a holistic representation for dynamic scenes rather than applying 3D-GS for each individual frame, and achieves real-time rendering under high resolutions, 82 FPS at an 800x800 resolution on an RTX 3090 GPU while maintaining comparable or better quality than previous state- of-the-art methods.
- 1,273
- Adversarial Examples for Semantic Segmentation and Object Detection1,057
This paper proposes a novel algorithm named Dense Adversary Generation (DAG), which applies to the state-of-the-art networks for segmentation and detection, and finds that the adversarial perturbations can be transferred across networks with different training data, based on different architectures, and even for different recognition tasks.
- Genetic CNN942
The core idea is to propose an encoding method to represent each network structure in a fixed-length binary string to efficiently explore this large search space.
- Genetic CNN942
The core idea is to propose an encoding method to represent each network structure in a fixed-length binary string to efficiently explore this large search space.
- Genetic CNN942
The core idea is to propose an encoding method to represent each network structure in a fixed-length binary string to efficiently explore this large search space.
- Conformer: Local Features Coupling Global Representations for Visual Recognition827
Experiments show that Conformer, under the comparable parameter complexity, outperforms the visual transformer (DeiT-B) by 2.3% on ImageNet and on MSCOCO, it outperforms ResNet-101 by 3.7% and 3.6% mAPs for object detection and instance segmentation, respectively, demonstrating the great potential to be a general backbone network.
- Progressive Differentiable Architecture Search: Bridging the Depth Gap Between Search and Evaluation751
This paper presents an efficient algorithm which allows the depth of searched architectures to grow gradually during the training procedure, and solves two issues, namely, heavier computational overheads and weaker search stability, which are solved using search space approximation and regularization.
- Fast Dynamic Radiance Fields with Time-Aware Neural Voxels492
This framework significantly accelerates the optimization of dynamic radiance fields while maintaining high rendering quality and completes training with only 8 minutes and 8-MB storage cost while showing similar or even better rendering performance than previous dynamic NeRF methods.
- 376
- SIXray: A Large-Scale Security Inspection X-Ray Benchmark for Prohibited Item Discovery in Overlapping Images362
A large-scale dataset and establish a baseline for prohibited item discovery in Security Inspection X-ray images, in which 6 classes of 8,929 prohibited items are manually annotated, and proposes an approach named class-balanced hierarchical refinement (CHR) to deal with these difficulties.
- GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models339
This paper attempts to bridge the power from the two types of diffusion models via the recent explicit and efficient 3D Gaussian splatting representation via a fast 3D object gener-ation framework, named as GaussianDreamer, where the 3D diffusion model provides priors for initial-ization and the 2D diffusion model enriches the geometry and appearance.
- Visformer: The Vision-friendly Transformer328
A new architecture named Visformer is proposed, which is abbreviated from the ‘Vision-friendly Transformer’, which outperforms both the Transformer-based and convolution-based models in terms of ImageNet classification accuracy, and the advantage becomes more significant when the model complexity is lower or the training set is smaller.
- A Fixed-Point Model for Pancreas Segmentation in Abdominal CT Scans317
This paper forms a fixed-point model which uses a predicted segmentation mask to shrink the input region and outperform the state-of-the-art by more than \(4\%\), measured by the average Dice-Sorensen Coefficient (DSC).
- Unsupervised Person Re-Identification via Softened Similarity Learning307
The iterative training mechanism is followed but clustering is discarded, since it incurs loss from hard quantization, yet its only product, image-level similarity, can be easily replaced by pairwise computation and a softened classification task.
- Pangu-Weather: A 3D High-Resolution Model for Fast and Accurate Global Weather Forecast301
Pangu-Weather not only ends the debate on whether AI-based methods can surpass conventional NWP methods, but also reveals novel directions for improving deep learning weather forecast systems.
- Attention-Guided Unified Network for Panoptic Segmentation293
The underlying relationship between FG objects and BG contents is revealed, in particular, FG objects provide complementary cues to assist BG understanding, and the Attention-guided Unified Network (AUNet) is named, a unified framework with two branches for FG and BG segmentation simultaneously.
- DisturbLabel: Regularizing CNN on the Loss Layer266
An extremely simple algorithm which randomly replaces a part of labels as incorrect values in each iteration, which prevents the network training from over-fitting by implicitly averaging over exponentially many networks which are trained with different label sets.
- 253
- 245
- 231
- QA-LoRA: Quantization-Aware Low-Rank Adaptation of Large Language Models227
A quantization-aware low-rank adaptation (QA-LoRA) algorithm that equips the original LoRA with two-fold abilities: during fine-tuning, the LLM's weights are quantized (e.g., into INT4) to reduce time and memory usage; and after fine- Tuning,the LLM and auxiliary weights are naturally integrated into a quantized model without loss of accuracy.
- Pruning from Scratch225
This work finds that pre-training an over-parameterized model is not necessary for obtaining the target pruned structure, and empirically shows that more diverse pruned structures can be directly pruned from randomly initialized weights, including potential models with better performance.
- GaussianEditor: Editing 3D Gaussians Delicately with Text Instructions220
A systematic framework to edit 3D scenes delicately via 3D Gaussians with text instructions, which can achieve more delicate and precise editing of 3D scenes than previous methods while enjoying much faster training speed, i.e. within 20 minutes on a single V100 GPU.
- Adaptive Keyframe Sampling for Long Video Understanding217
Experiments on two long video understanding benchmarks validate that AKS improves video QA accuracy (beyond strong baselines) upon selecting informative keyframes, and reveals the importance of information pre-filtering in video-based MLLMs.
- Bottom-Up Temporal Action Localization with Mutual Regularization213
Two regularization terms are introduced to mutually regularize the learning procedure for temporal action localization, and the Intra-phase Consistency (IntraC) regularization is proposed to make the predictions verified inside each phase; and the Inter- phase Consistencies (InterC)Regularization is suggested to keep consistency between these phases.
- Rethinking the Distribution Gap of Person Re-identification with Camera-Based Batch Normalization191
This paper rethinks the working mechanism of conventional ReID approaches and puts forward a new solution, with an effective operator named Camera-based Batch Normalization (CBN), that forces the image data of all cameras to fall onto the same subspace, so that the distribution gap between any camera pair is largely shrunk.
- 187
- Adversarial Attacks Beyond the Image Space165
Though image-space adversaries can be interpreted as per-pixel albedo change, it is verified that they cannot be well explained along these physically meaningful dimensions, which often have a non-local effect.
- A Survey on Label-Efficient Deep Image Segmentation: Bridging the Gap Between Weak Supervision and Dense Prediction159
A taxonomy is developed to organize the existing label-efficient image segmentation methods according to the supervision provided by different types of weak labels and discusses an important question: how to bridge the gap between weak supervision and dense prediction.
- 153
- Training Deep Neural Networks in Generations: A More Tolerant Teacher Educates Better Students152
Although the teacher network behaves less powerful, the students show a persistent ability growth and eventually achieve higher classification accuracies than other competitors and model ensemble and transfer feature extraction also verify the effectiveness of the approach.
- PC-DARTS: Partial Channel Connections for Memory-Efficient Differentiable Architecture Search.150
This paper presents a novel approach, namely, Partially-Connected DARTS, by sampling a small part of super-network to reduce the redundancy in exploring the network space, thereby performing a more efficient search without comprising the performance.
- Iterative Reorganization With Weak Spatial Constraints: Solving Arbitrary Jigsaw Puzzles for Unsupervised Representation Learning149
This paper presents a novel approach which applies to jigsaw puzzles with an arbitrary grid size and dimensionality and provides a fundamental and generalized principle, that weaker cues are easier to be learned in an unsupervised manner and also transfer better.
- 148
- Creating Something From Nothing: Unsupervised Knowledge Distillation for Cross-Modal Hashing139
This paper proposes a novel approach that enables guiding a supervised method using outputs produced by an unsupervised method, and makes use of teacher-student optimization for propagating knowledge.
- Image Classification and Retrieval are ONE138
It is demonstrated that the essentials of image classification and retrieval are the same, since both tasks could be tackled by measuring the similarity between images, and a unified algorithm is proposed, which only involves manual object definition, regional description and nearest-neighbor search.
- Weight-Sharing Neural Architecture Search: A Battle to Shrink the Optimization Gap132
A literature review on the application of NAS to computer vision problems is provided and existing approaches are summarized into several categories according to their efforts in bridging the gap.
- Semantic-Aware Knowledge Preservation for Zero-Shot Sketch-Based Image Retrieval132
This paper designs an approach named Semantic-Aware Knowledge prEservation (SAKE), which fine-tunes the pre-trained model in an economical way and leverages semantic information, e.g., inter-class relationship, to achieve the goal of knowledge preservation.
- Social Adaptive Module for Weakly-Supervised Group Activity Recognition121
This paper presents a new task named weakly-supervised group activity recognition (GAR) which differs from conventional GAR tasks in that only video-level labels are available, yet the important persons within each frame are not provided even in the training data.
- Corner Proposal Network for Anchor-Free, Two-Stage Object Detection121
A novel anchor-free, two-stage framework which first extracts a number of object proposals by finding potential corner keypoint combinations and then assigns a class label to each proposal by a standalone classification stage is proposed, which is competitive among state-of-the-art object detection methods.
- Multi-scale Coarse-to-Fine Segmentation for Screening Pancreatic Ductal Adenocarcinoma114
An intuitive approach of detecting pancreatic ductal adenocarcinoma (PDAC), the most common type of pancreatic cancer, by checking abdominal CT scans is proposed, and multi-scale segmentation-for-classification, which classifies volumes by checking if at least a sufficient number of voxels is segmented as tumors, is named.
- CenterNet++ for Object Detection112
It is demonstrated that bottom-up approaches show competitive performance compared with top-down approaches and have higher recall rates and a real-time CenterNet model, an anchor-free detector that achieves a good trade-off between accuracy and speed.
- 109
- Progressive DARTS: Bridging the Optimization Gap for NAS in the Wild109
A progressive method that gradually increases the network depth during the search stage, which leads to the Progressive DARTS (P-DARTS) algorithm, which achieves improved performance on both the proxy dataset (CIFAR10) and a few target problems.
- Deep Supervision for Pancreatic Cyst Segmentation in Abdominal CT Scans104
Inspired by the high relevance between the location of a pancreas and its cystic region, extra deep supervision is introduced into the segmentation network, so that cyst segmentation can be improved with the help of relatively easier Pancreas segmentation.
- GaussianObject: High-Quality 3D Object Reconstruction from Four Views with Gaussian Splatting101
GaussianObject, a framework to represent and render the 3D object with Gaussian splatting that achieves high rendering quality with only 4 input images, is proposed and evaluated on several challenging datasets, achieving superior performance from only four views and significantly outperforming previous SOTA methods.
- Attention-based Pyramid Aggregation Network for Visual Place Recognition90
An Attention-based Pyramid Aggregation Network (APANet) is presented, which is trained in an end-to-end manner for place recognition and significantly improves the widely used PCA whitening by reasonably limiting the impact of over-counting.
- Object Recognition with and without Objects90
This paper trains deep neural networks on the foreground (object) and background (context) regions of images respectively and shows that useful visual hints can be explicitly learned separately and then combined to achieve higher performance, which verifies the advantages of the proposed framework.
- MSG-Transformer: Exchanging Local Spatial Information by Manipulating Messenger Tokens89
This paper aims to alleviate the conflict between efficiency and flexibility, for which it proposes a specialized token for each region that serves as a messenger (MSG) that is integrated into a multi-scale architecture named MSG-Transformer.
- 89
- 87
- 86
- UnrealPerson: An Adaptive Pipeline towards Costless Person Re-identification84
UnrealPerson is presented, a novel pipeline that makes full use of unreal image data to decrease the costs in both the training and deployment stages and is a system that can generate synthesized images of high-quality and from controllable distributions.
- 82
- 81
- 80
- 80
- Segment Anything in 3D with Radiance Fields78
The research reveals a potential methodology to lift the ability of a 2D segmentation model to 3D and designs an efficient solution, leveraging the radiance field as a cheap and off-the-shelf prior that connects multi-view 2D images to the 3D space.
- Bridging the Gap Between 2D and 3D Organ Segmentation with Volumetric Fusion Net78
An alternative framework is presented, which trains 2D networks on different viewpoints for segmentation, and builds a 3D Volumetric Fusion Net (VFN) to fuse the 2D segmentation results, which is relatively shallow and contains much fewer parameters than most 3D networks.
- Visual Tuning76
This survey characterizes a large and thoughtful selection of recent works, providing a systematic and comprehensive overview of existing work and models, and categorizes recent visual tuning techniques into five groups: fine-tuning, prompt tuning, adapter tuning, parameter tuning, and remapping tuning.
- 74
- Understanding and Mitigating Overfitting in Prompt Tuning for Vision-Language Models71
Subspace Prompt Tuning (Sub PT) is proposed to project the gradients in back-propagation onto the low-rank subspace spanned by the early-stage gradient flow eigenvectors during the entire training process and successfully eliminate the overfitting problem.
- 71
- Circumventing Outliers of AutoAugment with Knowledge Distillation68
It is revealed that AutoAugment may remove part of discriminative information from the training image and so insisting on the ground-truth label is no longer the best option, and knowledge distillation is made use that refers to the output of a teacher model to guide network training.
- CRAVES: Controlling Robotic Arm With a Vision-Based Economic System66
This work designs a semi-supervised approach to low-cost arms on which no sensors are equipped and thus all decisions are made upon visual recognition, e.g., real-time 3D pose estimation and applies it to real-world images after domain adaptation.
- 59
- 56
- SORT: Second-Order Response Transform for Visual Recognition56
A novel approach named Second-Order Response Transform (SORT), which appends element-wise product transform to the linear sum of a two-branch network module, which augments the family of transform operations and increases the nonlinearity of the network.
- Fast Non-Local Neural Networks with Spectral Residual Learning52
Spectral residual learning is proposed, a novel network architectural design for achieving fully global receptive field and its equivalence to conducting residual learning in some spectral domain is shown and performance improvement by large margins is shown.
- InterActive: Inter-Layer Activeness Propagation51
InterActive is presented, a novel algorithm which computes the activeness of neurons and network connections, and achieves state-of-the-art classification performance on a wide range of image datasets.
- HiViT: Hierarchical Vision Transformer Meets Masked Image Modeling50
A new design of hierarchical vision transformers named HiViT (short for Hierarchical ViT) is offered that enjoys both high efficiency and good performance in MIM and is able to remove the unnecessary"local inter-unit operations", deriving structurally simple hierarchical visiontransformers in which mask-units can be serialized like plain vision transformer.
- 50
- 50
- USAGE: A Unified Seed Area Generation Paradigm for Weakly Supervised Semantic Segmentation49
This paper proposes a Unified optimization paradigm for Seed Area GEneration (USAGE) for both types of networks, in which the objective function to be optimized consists of two terms: a generation loss and a regularization loss, which ensures the consistency between the seed areas that are generated by self-adaptive network adjustment from different views.
- 49
- 48
- The FELIX Project: Deep Networks To Detect Pancreatic Neoplasms48
A suite of algorithms (named FELIX) that can recognize pancreatic lesions from CT images without human input are described that may be able to assist radiologists in identifying pancreatic cancers earlier, when surgery and other treatments offer more hope for long-term survival.
- 47
- Polar Relative Positional Encoding for Video-Language Segmentation47
A novel Polar Relative Positional Encoding (PRPE) mechanism that represents spatial relations in a ``linguistic'' way, i.e., in terms of direction and range, which outperforms previous best method by a large margin.
- 46
- Fast and accurate near-duplicate image search with affinity propagation on the ImageWeb46
This paper constructs ImageWeb, a sparse graph consisting of all the images in the database, in which two images are connected if and only if one is ranked among the top of another’s initial search result, and uses HITS, a query-dependent algorithm to re-rank the images according to the affinity values.
- Appending Adversarial Frames for Universal Video Attack45
A new attack approach is proposed which only adds perturbations on the newly-added frames of a modified video and shows its high success rates in attacking six state-of-the-art video classification networks, as well as its universality, i.e., transferring well across videos and models.
- Fast-iTPN: Integrally Pre-Trained Transformer Pyramid Network With Token Migration43
Fast-iTPN can accelerate the inference procedure by up to up to 70% with negligible performance loss, demonstrating the potential to be a powerful backbone for downstream vision tasks.
- 43
- 42
- MagDR: Mask-guided Detection and Reconstruction for Defending Deepfakes42
MagDR, a mask-guided detection and reconstruction pipeline for defending deepfakes from adversarial attacks, shows promising performance in defending both black-box and white-box attacks.
- Domain-Agnostic Prior for Transfer Semantic Segmentation40
This paper presents a simple and effective mechanism that regularizes cross-domain representation learning with a domain-agnostic prior (DAP) that constrains the features extracted from source and target domains to align with a Domainagnostic space.
- Learnable Distribution Calibration for Few-Shot Class-Incremental Learning39
A learnable distribution calibration (LDC) approach, built upon a parameterized calibration unit (PCU), which initializes biased distributions for all classes based on classifier vectors (memory-free) and a single covariance matrix.
- Detecting Semantic Parts on Partially Occluded Objects38
Experiments show that the algorithm outperforms several competitors in semantic part detection when occlusions are present, and uses a simple voting method, based on log-likelihood ratio tests and spatial constraints, to combine the evidence of local cues.
- Discretization-aware architecture search37
Discretization-aware architecture search is presented, with the core idea being adding a loss term to push the super-network towards the configuration of desired topology, so that the accuracy loss brought by discretization is largely alleviate.
- 37
- Bridging the Gap Between 2D and 3D Organ Segmentation.37
An alternative framework is presented, which trains 2D networks on different viewpoints for segmentation, and builds a 3D Volumetric Fusion Net (VFN) to fuse the 2D segmentation results, which is relatively shallow and contains much fewer parameters than most 3D networks.
- Pancreas Segmentation in Abdominal CT Scan: A Coarse-to-Fine Approach.36
This paper trains two deep neural networks using different regions of the input volume and proposes a coarse-to-fine approach, which outperforms the state-of-the-art by more than 4% measured by the Dice-Sorensen Coefficient (DSC).
- Elastic Boundary Projection for 3D Medical Image Segmentation35
Elastic Boundary Projection allows boundary-based segmentation without cutting a 3D volume into slices or patches, which stands out from conventional 2D and 3D approaches.
- Visual concepts and compositional voting35
This work uses visual concepts as building blocks for a simple pattern theoretical model, which is significantly better than discriminative methods like SVM and deep networks trained specifically for semantic part detection, and shows that compositional voting outperforms even deep networks when the amount of occlusion becomes large.
- Towards Reversal-Invariant Image Representation35
A reversal-invariant version of SIFT descriptor named Max-SIFT, a generalized RIDE algorithm which can be applied to a large family of local descriptors which reveals consistent accuracy gain on various image classification tasks, including scene understanding, fine-grained object recognition, and large-scale visual recognition.
- 34
- 31
- 31
- ChatterBox: Multi-round Multimodal Referring and Grounding30
A new benchmark and an efficient vision-language model is presented, named ChatterBox, which outperforms existing models in MRG both quantitatively and qualitatively, paving a new path towards multimodal dialogue scenarios with complicated and precise interactions.
- Can Semantic Labels Assist Self-Supervised Visual Representation Learning?29
A new algorithm named Supervised Contrastive Adjustment in Neighborhood (SCAN) is presented that maximally prevents the semantic guidance from damaging the appearance feature embedding and reveals that semantic labels are useful in assisting self-supervised methods, opening a new direction for the community.
- Single Camera Training for Person Re-Identification29
This paper investigates ReID in an unexplored single-camera-training (SCT) setting, where each person in the training set appears in only one camera, and proposes a novel loss function named multi- camera negative loss (MCNL), a metric learning loss motivated by probability suggesting that in a multi-camera system, one image is more likely to be closer to the most similar negative sample in other cameras than to themost similarnegative sample in the same camera.
- 28
- Parameter Efficient Fine-Tuning via Cross Block Orchestration for Segment Anything Model27
This paper equips PEFT with a cross-block orchestration mechanism to enable the adaptation of the Segment Anything Model to various downstream scenarios, and introduces a novel inter-block communication module, which integrates a learnable relation matrix to facilitate communication among different coefficient sets of each PEFT block's parameter space.
- 27
- 27
- Reinforced Axial Refinement Network for Monocular 3D Object Detection27
The proposed framework, Reinforced Axial Refinement Network (RAR-Net), serves as a post-processing stage which can be freely integrated into existing monocular 3D detection methods, and improve the performance on the KITTI dataset with small extra computational costs.
- 27
- GaussianDreamerPro: Text to Manipulable 3D Gaussians with Highly Enhanced Quality26
This work proposes a novel framework named GaussianDreamerPro, which binds Gaussians to reasonable geometry, which evolves over the whole generation process and shows significantly enhanced details and quality compared with previous methods.
- 26
- 25
- 25
- 24
- Focus on Your Target: A Dual Teacher-Student Framework for Domain-adaptive Semantic Segmentation24
A novel dual teacher-student (DTS) framework is proposed and equip it with a bidirectional learning strategy, where the second teacher-student model learns to ‘Focus on Your Target’ while the first model is not affected.
- Beyond Masking: Demystifying Token-Based Pre-Training for Vision Transformers24
This paper designs five other learning objectives that follow the same procedure as MIM but degrade the input image in different ways, and achieves superior performance over MIM in a series of downstream recognition tasks without extra computational cost.
- Spatial pooling of heterogeneous features for image applications24
A novel framework with spatial pooling of heterogeneous features for mid-level representation of images and test the superior performance of the algorithm over state-of-the-art methods on several benchmark datasets for image classification and retrieval applications.
- 23
- Omni-GAN: On the Secrets of cGANs and Beyond23
Omni-GAN is presented, a variant of cGAN that reveals the devil in designing a proper discriminator for training the model, and can extrapolate low-resolution images to arbitrary resolution, even up to ×60+ higher resolution.
- Towards Multiple Black-boxes Attack via Adversarial Example Generation Network23
MBbA achieves the highest attack success rates in a single black-box attack scenario; the adversarial examples generated by MBbA can effectively improve the robustness of the attacked models and exhibit good transferability.
- 23
- 22
- 21
- 20
- Joint Shape Representation and Classification for Detecting PDAC20
A two-stage framework is proposed, which first segments the pancreas into a binary mask, then compresses the mask into a shape vector and performs abnormality classification, which shows promise for clinical applications.
- 20
- Visual Recognition by Request19
A new paradigm named visual recognition by request is established, which allows for learning complicated whole-part hierarchies from highly incomplete annotations and inserting new concepts with minimal efforts, and establishes a solid baseline by integrating language-driven recognition into recent semantic and instance segmentation methods.
- 18
- Multi-scale Spatially-Asymmetric Recalibration for Image Classification18
This paper introduces a recalibration process, which refers to the surrounding region of each neuron, computes an importance value and multiplies it to the original neural response, and designs a weighting scheme which is asymmetric in the spatial domain.
- Fast Nearest Neighbor Search in the Hamming Space18
This paper generalizes the algorithm to the Hamming space with an alternative version of k-means clustering and achieves competitive performance compared to the state-of-the-art methods, i.e., MIH and FLANN, in the aspects of search precision, accessed data volume and average querying time.
- Incorporating visual adjectives for image classification18
This paper proposes a simple framework to integrate various types of adjectives, i.e., color, shape and location, for more powerful image representation, and shows that this approach achieves superior classification accuracy with reasonable computational overheads.
- Fitting the Search Space of Weight-sharing NAS with Graph Convolutional Networks16
This paper trains a graph convolutional network to fit the performance of sampled sub-networks so that the impact of random errors becomes minimal and achieves a higher rank correlation coefficient in the selected set of candidates, which consequently leads to better performance of the final architecture.
- Searching Towards Class-Aware Generators for Conditional Generative Adversarial Networks15
This paper presents an idea that adopts neural architecture search to find a class-aware architecture for each class, and finds two intriguing phenomena that are inspirational to craft cGANs by hand.
- 15
- Universal-to-Specific Framework for Complex Action Recognition15
An effective universal-to-specific (U2S) framework for complex action recognition is proposed, and the relationship between different classes is visualize, showing that U2S indeed improves the discriminability of learned features.
- 15
- Network Adjustment: Channel Search Guided by FLOPs Utilization Ratio15
A new framework named network adjustment is presented, which considers network accuracy as a function ofFLOPs, so that under each network configuration, one can estimate the FLOPs utilization ratio (FUR) for each layer and use it to determine whether to increase or decrease the number of channels on the layer.
- LiftImage3D: Lifting Any Single Image to 3D Gaussians with Video Generation Priors14
This work designs an articulated trajectory strategy to generate video frames, which decomposes video sequences with large camera motions into ones with controllable small motions, and proposes a distortion-aware 3D Gaussian splatting representation, which can learn independent distortions between frames and output undistorted canonical Gaussians.
- Heterogeneous Contrastive Learning: Encoding Spatial Information for Compact Visual Representations14
This paper presents heterogeneous contrastive learning (HCL), an effective approach that adds spatial information to the encoding stage to alleviate the learning inconsistency between the contrastive objective and strong data augmentation operations.
- 14
- 14
- 14
- 13
- 13
- Consensus Synergizes With Memory: A Simple Approach for Anomaly Segmentation in Urban Scenes12
Experimental results on several urban scene anomaly segmentation datasets demonstrate that CosMe outperforms previous approaches by a significant margin.
- Hierarchical Semantic Aggregation for Contrastive Representation Learning.12
This paper tackles the representation inefficiency of contrastive learning and proposes a hierarchical training strategy to explicitly model the invariance to semantic similar images in a bottom-up way and produces more discriminative representation on several unsupervised benchmarks.
- 11
- DATA: Domain-Aware and Task-Aware Self-supervised Learning11
This paper presents DATA, a simple yet effective NAS approach specialized for SSL that provides Domain-Aware and Task- aware pre-training and proposes a flexible searching mechanism compatible with SSL that enables finding networks of different computation costs, for various downstream vision tasks and data domains without explicit metric provided.
- 11
- Geometric Neural Phrase Pooling: Modeling the Spatial Co-occurrence of Neurons11
The Geometric Neural Phrase Pooling algorithm is proposed, which acts as a new type of hidden layer, which punishes the isolated neuron responses after convolution, and can be inserted into a CNN model with little extra computational overhead.
- 11
- 10
- AlignZeg: Mitigating Objective Misalignment for Zero-Shot Semantic Segmentation10
AlignZeg markedly enhances zero-shot semantic segmentation, as shown by an average 3.8% increase in hIoU, primarily attributed to a 7.1% improvement in identifying unseen classes, and it is validated that the improvement comes from alleviating the objective misalignment issue.
- Semantic Part Detection via Matching: Learning to Generalize to Novel Viewpoints From Limited Training Data10
This paper presents an approach which can learn from a small annotated dataset containing a limited range of viewpoints and generalize to detect semantic parts for a much largerrange of viewpoints.
- 10
- 9
- Towards AGI in Computer Vision: Lessons Learned from GPT and Large Language Models9
It is pointed out that the essential weakness of CV lies in lacking a paradigm to learn from environments, yet NLP has accomplished the task in the text world and is still far from a system like GPT that naturally integrates all tasks.
- 9
- Thickened 2D Networks for 3D Medical Image Segmentation.9
A novel approach which thickens the input of a 2D network, so that the model is expected to enjoy both the stability and efficiency of 2D networks as well as the ability of 3D networks in modeling volumetric contexts.
- DeepVoting: An Explainable Framework for Semantic Part Detection under Partial Occlusion.9
A novel detection framework, named DeepVoting, which accumulates local visual cues, called visual concepts (VC), to locate the semantic parts of an object, and shows significantly better performance on semantic part detection in occlusion scenarios, compared with Faster-RCNN.
- Domain-Agnostic Priors for Semantic Segmentation Under Unsupervised Domain Adaptation and Domain Generalization8
This paper presents a novel framework that forces visual features to align with domain-agnostic priors (DAP), and studies two kinds of priors, language-guided embedding and class-level relationship, which work sufficiently well for UDA, while class-level priors serve as useful complements for DG.
- VoxSeP: semi-positive voxels assist self-supervised 3D medical segmentation8
This paper investigates self-supervised pre-training, making use of unlabeled medical data to initialize 3D segmentation networks, and presents a novel proxy task that takes advantage of the property of human body similarity in medical scans, and defines the sub-volumes from the same position of different cases as Semi-Positive samples.
- ChatterBox: Multimodal Referring and Grounding with Chain-of-Questions7
Experiments show that ChatterBox demonstrates superiority in MCQ both quantitatively and qualitatively, paving a new path towards multimodal dialogue scenarios with logical interactions.
- Pipeline MoE: A Flexible MoE Implementation with Pipeline Parallelism7
A novel MoE architecture called Pipeline MoE (PPMoE) is proposed that builds expert parallel incorporating with tensor parallel and replaces communication-intensive all-to-all dispatching and gathering with a simple tensor index slicing and inner-node all-reduce.
- 7
- 7
- G2C: A Generator-to-Classifier Framework Integrating Multi-Stained Visual Cues for Pathological Glomerulus Classification7
The approach, named generator-to-classifier (G2C), is a two-stage framework that combines visual cues from different stains for prediction, and boosts classification performance, suggesting more effective visual cues are extracted in an automatic way.
- 7
- 6
- 6
- Identity-Enhanced Network for Facial Expression Recognition6
A novel identity-enhanced network (IDEnNet) is presented to eliminate the negative impact of identity factor and focus on recognizing facial expressions, and achieves the best or comparable state-of-the-art on all three datasets.
- Phase Collaborative Network for Multi-Phase Medical Imaging Segmentation.6
This paper proposes Phase Collaborative Network (PCN), an end-to-end network which contains both generative and discriminative modules to formulate phase- to-phase relations and data-to -label relations, respectively and empirically verify that the accuracy gain comes from the collaboration between phases.
- Text-Animator: Controllable Visual Text Video Generation5
Quantitative and qualitative experimental results demonstrate the superiority of the proposed Text-Animator approach to the accuracy of generated visual text over state-of-the-art video generation methods.
- 5
- 5
- ClawMachine: Learning to Fetch Visual Tokens for Referential Comprehension4
ClawMachine is presented, offering a new methodology that notates an entity directly using the visual tokens without additional syntax to unify the prompt and answer of visual referential tasks without additional syntax.
- Exploring the diversity and invariance in yourself for visual pre-training task4
This paper introduces a simple but effective mechanism, called Exploring the Diversity and Invariance in Yourself E-DIY, which maximally extracts the multi-grained visual information inside each image.
- Network Adjustment: Channel and Block Search Guided by Resource Utilization Ratio4
A novel framework named network adjustment is presented which considers accuracy as a function of the computational resource (e.g., FLOPs or parameters) so that architecture design becomes an optimization problem and can be solved with the gradient-based optimization method.
- Cornerformer: Purifying Instances for Corner-Based Detectors4
This paper builds a Corner Transformer Encoder in a 2D-form to enhance the information aggregated by corner keypoints, offering stronger features for the pull-push loss to distinguish instances from each other and designs an Attenuation-Auto-Adjusted NMS to maximally leverage the semantic outputs and avoid true objects from being removed.
- Scalable NAS with factorizable architectural parameters4
This paper presents a scalable algorithm by factorizing a large set of candidate operators into smaller subspaces that allows to search for effective activation functions along with the regular operators including convolution, pooling, skip-connect, etc.
- 4
- Vibration-Based Uncertainty Estimation for Learning from Limited Supervision3
A novel approach is presented that measures the uncertainty from the vibration of sequential data, so that deep neural networks can make use of the results for learning from limited supervision.
- 3
- 3
- Towards Accurate Task Accomplishment with Low-Cost Robotic Arms3
This work designs a semi-supervised approach to low-cost arms on which no sensors are equipped and thus all decisions are made upon visual recognition, e.g., real-time 3D pose estimation, and applies an iterative algorithm for optimization.
- Saliency Transformation Network: Incorporating Multi-stage Visual Cues for Pancreas Segmentation.3
A Saliency Transformation Network is presented, which contains a trainable saliency transformation module that computes spatial weights from the coarse-scaling segmentation score map, and applies them to the fine-scaled input image.
- DiffCrack: A semantic–structural controllable framework for crack image generation in complex scenes2
- P2Object: Single Point Supervised Object Detection and Instance Segmentation2
This paper introduces Point-to-Box Network (P2BNet), which constructs balanced instance-level proposal bags by generating proposals in an anchor-like way and refining the proposals in a coarse-to-fine paradigm, and largely surpasses the previous methods in terms of the mean average precision.
- 2
- One-Bit Supervision for Image Classification: Problem, Solution, and Beyond2
Theoretical analysis shows that one-bit annotation is more efficient than full-bit annotation in most cases and gives the conditions of combining the authors' approach with active learning, and further integrates the one-bit supervision framework into the self-supervised learning algorithm, which yields an even more efficient training schedule.
- 2
- 1
- 1
- Finding the Host from the Lesion by Iteratively Mining the Registration Graph1
This paper investigates an interesting problem that finds the host organ of a lesion without actually labeling the organ, and boosts the accuracy of tumor segmentation to a considerable degree, $3%$, which even surpasses the model trained with ground-truth of both organ and tumor.
- 1
- Constraining Temporal Relationship for Action Localization.1
This paper delves deep into the mechanism of temporal action localization, and argues that existing approaches mostly ignored the potential relationship of these curves, and results in low quality of action proposals.
- Disassembling the Dataset: A Camera Alignment Mechanism for Multiple Tasks in Person Re-identification1
A fine-grained distribution alignment formulation, which disassembles the dataset and aligns all training and testing cameras, and devise the Camera-based Batch Normalization, which is easy for integration and nearly cost-free for existing ReID methods.
- 1
- 2D-Based Coarse-to-Fine Approaches for Small Target Segmentation in Abdominal CT Scans1
This chapter proposes two coarse-to-fine mechanisms which use prediction from the first (coarse) stage to shrink the input region for the second (fine) stage and inserts a saliency transformation module between the two stages.
- 1
- Towards multi-modality invariance: A study in visual representation1
This chapter presents a comprehensive study in extracting invariant compact visual features, i.e., visual features that resist the changes in position, viewpoints, geometry, etc., serving as a basic of understanding video signals and assisting other modalities such as audio signals.
- 1
- 1
- –
- –
- –
- Exploring Complicated Search Spaces With Interleaving-Free Sampling–
This study explores the search algorithm upon a complicated search space with long-distance connections and shows that existing weight-sharing search algorithms fail due to the existence of interleaved connections, and presents a simple-yet-effective algorithm, termed interleaving-free neural architecture search (IF-NAS).
- –
- –
- –
- –
- –
- Progressive Recurrent Learning for Visual Recognition.–
This paper applies progressive recurrent learning to two challenging visual recognition tasks, namely, object localization and semantic segmentation, and demonstrates consistent accuracy gain compared to the baseline training strategy.
- –
- –
- –
- –
- –
Publication data from OpenAlex; 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.