ICCV 2023 Papers with AI Summaries
IEEE/CVF International Conference on Computer Vision · 2156 papers
→ ICCV 2023 papers with code (743)
Each paper below shows an AI-generated one-line summary. Get the full 6-part summary (innovation, method, data, results, limitations) and search all 2156 ICCV 2023 papers by keyword, author or institution —
free trial on arXivSub.
2D-3D Interlaced Transformer for Point Cloud Segmentation with Scene-Level Supervision
Cheng-Kun Yang (National Taiwan University), Yen-Yu Lin (National Yang Ming Chiao Tung University)
SegmentationTransformerContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: This paper proposes a Multi-modal Interleaved Transformer (MIT) that integrates 2D multi-view images with 3D point cloud features under weak supervision conditions, achieving point cloud segmentation using only scene-level category labels.
2D3D-MATR: 2D-3D Matching Transformer for Detection-Free Registration Between Images and Point Clouds
Minhao Li (National University of Defense Technology), Kai Xu (National University of Defense Technology)
TransformerImagePoint Cloud
🎯 What it does: A 2D-3D matching Transformer (2D3D-MATR) for free detection is proposed, implementing precise registration between images and point clouds through a coarse-to-fine matching pipeline.
360VOT: A New Benchmark Dataset for Omnidirectional Visual Object Tracking
Huajian Huang (Hong Kong University of Science and Technology), Sai-Kit Yeung (Hong Kong University of Science and Technology)
Object TrackingTransformerVideoBenchmark
🎯 What it does: This paper presents the 360VOT dataset, focusing on visual object tracking research for 360° videos generated by panoramic cameras.
3D Distillation: Improving Self-Supervised Monocular Depth Estimation on Reflective Surfaces
Xuepeng Shi (Imperial College London), Mohsen Ghafoorian (Qualcomm)
Depth EstimationKnowledge DistillationPoint Cloud
🎯 What it does: By combining self-supervised monocular depth estimation with a 3D distillation method, pseudo-labels are generated for reflective surfaces using the projected depth from multi-view reconstruction, thereby improving depth prediction accuracy.
3D Human Mesh Recovery with Sequentially Global Rotation Estimation
Dongkai Wang (Peking University), Shiliang Zhang (Peking University)
Pose EstimationConvolutional Neural NetworkSupervised Fine-TuningImageMesh
🎯 What it does: A Sequentially Global Rotation Estimation (SGRE) method is proposed for 3D human mesh recovery from monocular RGB images, directly predicting the global rotation matrices of each joint.
3D Implicit Transporter for Temporally Consistent Keypoint Discovery
Chengliang Zhong (Tsinghua University), Jian Zhao (Tsinghua University)
Object DetectionRobotic IntelligencePoint Cloud
🎯 What it does: A self-supervised 3D Implicit Transporter is proposed, capable of discovering spatiotemporally consistent key points from continuous point cloud sequences, and utilizing these key points for goal-driven 3D object manipulation.
3D Instance Segmentation via Enhanced Spatial and Semantic Supervision
Salwa Al Khatib (Mohamed Bin Zayed University of Artificial Intelligence), Fahad Shahbaz Khan (Linkoping University)
Object DetectionSegmentationConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: A 3D instance segmentation framework is proposed that integrates a sparse convolutional encoder with a Transformer decoder, incorporating spatial and semantic supervision in the encoder to enhance mask prediction accuracy using voxel coordinates.
3D Motion Magnification: Visualizing Subtle Motions from Time-Varying Radiance Fields
Brandon Y. Feng (University of Maryland), Jia-bin Huang
GenerationData SynthesisNeural Radiance FieldVideo
🎯 What it does: Using NeRF to achieve 3D motion magnification allows for the separation of scene motion from camera motion when the camera moves, and magnifies subtle movements from any viewpoint.
3D Neural Embedding Likelihood: Probabilistic Inverse Graphics for Robust 6D Pose Estimation
Guangyao Zhou (Google DeepMind), Vikash K. Mansinghka (Massachusetts Institute of Technology)
Pose EstimationImage
🎯 What it does: Proposes 3D Neural Embedding Likelihood (3DNEL), a 6D object pose estimation framework based on probabilistic inverse graphics.
3D Segmentation of Humans in Point Clouds with Synthetic Data
Ayça Takmaz (ETH Zurich), Siyu Tang (RWTH Aachen University)
SegmentationData SynthesisTransformerPoint Cloud
🎯 What it does: We propose Human3D, an end-to-end Transformer architecture that can directly perform human instance segmentation and multi-human part segmentation on point clouds, pre-trained with synthetic training data and then fine-tuned on real data.
3D Semantic Subspace Traverser: Empowering 3D Generative Model with Shape Editing Capability
Ruowei Wang (Sichuan University), Qijun Zhao (Sichuan University)
GenerationData SynthesisGenerative Adversarial NetworkPoint CloudMesh
🎯 What it does: A 3D Semantic Subspace Traverser is proposed, which can achieve 3D shape generation and semantic editing under implicit function representation.
3D VR Sketch Guided 3D Shape Prototyping and Exploration
Ling Luo (University of Surrey), Yulia Gryaditskaya (University of Surrey)
GenerationFlow-based ModelAuto EncoderPoint Cloud
🎯 What it does: This study investigates how to generate diverse 3D shapes that are consistent with the structure of sparse, abstract 3D VR hand-drawn sketches.
3D-aware Blending with Generative NeRFs
Hyunsu Kim (NAVER AI Lab), Jun-Yan Zhu (Carnegie Mellon University)
Image HarmonizationGenerationPose EstimationConvolutional Neural NetworkNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: This paper studies a 3D perception image blending method based on generative NeRF, which can achieve natural and seamless fusion between unaligned images.
3D-Aware Generative Model for Improved Side-View Image Synthesis
Kyungmin Jo (KAIST), Sunghyun Cho (POSTECH)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A new 3D GAN training method called SideGAN is proposed for generating 3D consistent images that are independent of camera pose and have high facial image quality.
3D-aware Image Generation using 2D Diffusion Models
Jianfeng Xiang (Tsinghua University), Xin Tong (Microsoft Research Asia)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A 3D perception image generation method based on a 2D diffusion model is proposed, which decomposes the 3D generation task into the generation of a multi-view 2D image set, achieving the generation of high-quality images from an unstructured single image set from multiple angles.
3D-Aware Neural Body Fitting for Occlusion Robust 3D Human Pose Estimation
Yi Zhang (Johns Hopkins University), Alan Yuille (University of Freiburg)
Pose EstimationContrastive LearningImage
🎯 What it does: By constructing a voxel-based differentiable neural body volume model and performing analysis-synthesis in the feature space, 3D human pose estimation from a single image is achieved, with occlusion resistance.
3D-VisTA: Pre-trained Transformer for 3D Vision and Text Alignment
Ziyu Zhu (Tsinghua University), Qing Li (National Key Laboratory of General Artificial Intelligence, BIGAI)
Object DetectionSegmentationRetrievalTransformerVision Language ModelMultimodalityPoint Cloud
🎯 What it does: This paper proposes the 3D-VisTA model, which achieves 3D vision and text alignment through a unified self-attention Transformer, and constructs a large-scale ScanScribe dataset for pre-training.
3DHacker: Spectrum-based Decision Boundary Generation for Hard-label 3D Point Cloud Attack
Yunbo Tao (Huazhong University of Science and Technology), Wei Hu (Peking University)
ClassificationAdversarial AttackGraph Neural NetworkPoint Cloud
🎯 What it does: A black-box hard-label attack method for 3D point clouds, called 3DHacker, is proposed, which generates and iteratively optimizes high-quality adversarial samples using spectral domain decision boundaries.
3DHumanGAN: 3D-Aware Human Image Generation with 3D Pose Mapping
Zhuoqian Yang (École Polytechnique Fédérale de Lausanne), Bo Dai (Shanghai Artificial Intelligence Laboratory)
GenerationData SynthesisPose EstimationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A 3D-aware GAN generator has been developed that can synthesize full-body portraits under given poses and viewpoints while ensuring appearance consistency.
3DMiner: Discovering Shapes from Large-Scale Unannotated Image Datasets
Ta-Ying Cheng (University of Oxford), Niki Trigoni (University of Oxford)
Object DetectionSegmentationRepresentation LearningTransformerContrastive LearningSimultaneous Localization and MappingImage
🎯 What it does: On a large-scale unlabeled image dataset, without the need for 3D annotations, camera information, or key points, the 3DMiner pipeline automatically mines and reconstructs 3D shapes;
3DMOTFormer: Graph Transformer for Online 3D Multi-Object Tracking
Shuxiao Ding (Mercedes-Benz AG), Juergen Gall (University of Bonn)
Object TrackingAutonomous DrivingGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: A 3D multi-object tracking framework called 3DMOTFormer is proposed, which is based on Edge-Augmented Graph Transformer and can perform online data association and trajectory prediction using only geometric information.
3DPPE: 3D Point Positional Encoding for Transformer-based Multi-Camera 3D Object Detection
Changyong Shu (Houmo AI), Yifan Liu (University of Adelaide)
Object DetectionDepth EstimationAutonomous DrivingKnowledge DistillationTransformerPoint Cloud
🎯 What it does: A 3D Point Pose Encoding (3DPPE) is proposed for Transformer-based multi-camera 3D object detection, utilizing depth estimation to project pixels into 3D space to obtain accurate point position information and embed features.
4D Myocardium Reconstruction with Decoupled Motion and Shape Model
Xiaohan Yuan (Southeast University), Yangang Wang (Southeast University)
SegmentationGenerationAuto EncoderImagePoint CloudMagnetic Resonance Imaging
🎯 What it does: A 4D myocardial reconstruction method based on implicit functions is proposed, which can predict the complete myocardial shape and its temporal evolution from sparse CMR slice point clouds.
4D Panoptic Segmentation as Invariant and Equivariant Field Prediction
Minghan Zhu (University of Michigan), Fatih Porikli (Qualcomm AI Research)
Object DetectionSegmentationAutonomous DrivingPoint CloudBenchmark
🎯 What it does: This paper proposes a rotation-equivariant neural network for 4D panoramic segmentation, aimed at identifying semantic categories and object instances on roads through LiDAR scanning, and assigning temporally consistent IDs to instances.
A 5-Point Minimal Solver for Event Camera Relative Motion Estimation
Ling Gao (ShanghaiTech University), Laurent Kneip (University of Zurich)
Pose EstimationOptimizationSimultaneous Localization and MappingImageVideoBenchmark
🎯 What it does: This paper proposes a minimum five-point solver based on event cameras, utilizing the event stripe (eventail) geometric model to jointly parameterize line features and the camera's linear velocity nonlinearly, achieving a preliminary estimate of the camera's linear motion;
A Benchmark for Chinese-English Scene Text Image Super-Resolution
Jianqi Ma (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
RestorationSuper ResolutionConvolutional Neural NetworkImageBenchmark
🎯 What it does: This paper constructs a realistic Chinese-English text image super-resolution benchmark dataset, Real-CE, and proposes a dual supervision learning method based on text edges to enhance the reconstruction quality of Chinese characters.
A Complete Recipe for Diffusion Generative Models
Kushagra Pandey (University of California), Stephan Mandt (University of California)
GenerationData SynthesisDiffusion modelScore-based ModelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes a complete design framework for constructing forward diffusion processes that satisfy a given target distribution, and introduces a new diffusion model called Phase Space Langevin Diffusion (PSLD) within this framework.
A Dynamic Dual-Processing Object Detection Framework Inspired by the Brain's Recognition Mechanism
Minying Zhang (Alibaba Group), Lulu Hu (Alibaba Group)
Object DetectionNeural Architecture SearchConvolutional Neural NetworkTransformerImage
🎯 What it does: A dynamic dual-processing (CNN+Transformer) object detection framework DDP is proposed, simulating the human brain's recognition mechanism. It utilizes a shared backbone network, a searchable dual-stream encoder, and a dynamically selected mask, allowing CNN and Transformer to work together in the encoder and decoder and adaptively select the decoder.
A Fast Unified System for 3D Object Detection and Tracking
Thomas Heitzinger (Vienna University of Technology), Martin Kampel (Vienna University of Technology)
Object DetectionObject TrackingPose EstimationConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: An end-to-end unified system named FUS3D has been designed and implemented, capable of achieving real-time 3D object detection and multi-object tracking on edge devices using only depth maps.
A Game of Bundle Adjustment - Learning Efficient Convergence
Amir Belder (Technion), Ayellet Tal (Technion)
Autonomous DrivingOptimizationReinforcement LearningImage
🎯 What it does: The damping factor λ of Bundle Adjustment (BA) is treated as a learnable dynamic variable, and suitable λ is selected in each iteration through reinforcement learning (RL) methods to accelerate the convergence of BA.
A Generalist Framework for Panoptic Segmentation of Images and Videos
Ting Chen (Google Deepmind), David J. Fleet (Google Deepmind)
Object TrackingSegmentationTransformerDiffusion modelImageVideo
🎯 What it does: A conditional discrete diffusion model Pix2SeqD is proposed for unified processing of panoramic segmentation tasks for images and videos.
A Good Student is Cooperative and Reliable: CNN-Transformer Collaborative Learning for Semantic Segmentation
Jinjing Zhu (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
SegmentationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: An online knowledge distillation framework is proposed, utilizing Heterogeneous Feature Distillation (HFD) and Bidirectional Selection Distillation (BSD) to achieve collaborative learning between CNN and ViT models for semantic segmentation tasks.
A Large-Scale Outdoor Multi-Modal Dataset and Benchmark for Novel View Synthesis and Implicit Scene Reconstruction
Chongshan Lu (Fudan University), Jiayuan Fan (Fudan University)
GenerationData SynthesisNeural Radiance FieldImageTextMultimodalityBenchmark
🎯 What it does: A large-scale outdoor multimodal aerial perspective dataset, OMMO, has been constructed, containing 33 real scenes, 14K camera calibration images, and text prompts, providing a new unified benchmark for NeRF tasks.
A Large-scale Study of Spatiotemporal Representation Learning with a New Benchmark on Action Recognition
Andong Deng (University of Central Florida), Chen Chen (University of Central Florida)
Domain AdaptationAnomaly DetectionRepresentation LearningConvolutional Neural NetworkTransformerSupervised Fine-TuningContrastive LearningVideoBenchmark
🎯 What it does: A new video action recognition benchmark BEAR is proposed, collecting action datasets from 18 sources across 5 domains (anomaly, gesture, daily, sports, teaching), and systematically evaluating 6 mainstream video models under various settings such as standard fine-tuning, few-shot fine-tuning, unsupervised domain adaptation, and zero-shot learning.
A Latent Space of Stochastic Diffusion Models for Zero-Shot Image Editing and Guidance
Chen Henry Wu (Carnegie Mellon University), Fernando De la Torre (Carnegie Mellon University)
Image TranslationGenerationDiffusion modelImageTextStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A latent space for random diffusion models is proposed, utilizing this space to achieve unpaired image translation, zero-shot image editing, and plug-in guidance.
A Low-Shot Object Counting Network With Iterative Prototype Adaptation
Nikola Đukić (University of Ljubljana), Matej Kristan (University of Ljubljana)
Object DetectionConvolutional Neural NetworkImageBenchmark
🎯 What it does: This paper proposes a low-sample object counting network (LOCA) that can accurately count the number of objects of any category in an image with only a few (or no) example boxes provided.
A Multidimensional Analysis of Social Biases in Vision Transformers
Jannik Brinkmann (University of Mannheim), Christian Bartelt (University of Mannheim)
RecognitionGenerationData SynthesisRepresentation LearningTransformerDiffusion modelContrastive LearningImage
🎯 What it does: This study investigates the causes and mitigation strategies of social bias in Vision Transformer (ViT) within self-supervised learning. It systematically evaluates the impact of factors such as training data, training objectives, model size, and input resolution on bias, and proposes a counterfactual augmentation strategy based on diffusion models.
A Parse-Then-Place Approach for Generating Graphic Layouts from Textual Descriptions
Jiawei Lin (Xi'an Jiaotong University), Dongmei Zhang (Microsoft Research Asia)
GenerationTransformerLarge Language ModelText
🎯 What it does: A text-based graphic layout generation method called Parse-Then-Place is proposed, which first parses the text into an intermediate representation (IR) and then generates the layout based on the IR;
A Retrospect to Multi-prompt Learning across Vision and Language
Ziliang Chen (Jinan University), Weiqi Luo (Jinan University)
ClassificationObject DetectionSegmentationRetrievalDomain AdaptationMeta LearningTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes a multi-prompt learning framework based on an energy model (EMPL), which achieves unified optimization for cross-modal transfer and open vocabulary generalization of visual-language models by sampling to generate diverse prompts in the image feature and soft prompt space.
A Sentence Speaks a Thousand Images: Domain Generalization through Distilling CLIP with Language Guidance
Zeyi Huang (University of Wisconsin Madison), Yong Jae Lee (University of Wisconsin Madison)
Domain AdaptationKnowledge DistillationPrompt EngineeringVision Language ModelImage
🎯 What it does: A domain generalization method is proposed that utilizes text embedding alignment from a CLIP teacher for knowledge distillation, training small models to enhance robustness to unseen domains.
A Simple Framework for Open-Vocabulary Segmentation and Detection
Hao Zhang (Hong Kong University of Science and Technology), Lei Zhang (International Digital Economy Academy)
Object DetectionSegmentationTransformerVision Language ModelImage
🎯 What it does: OpenSeeD is proposed, a unified framework capable of simultaneously handling open vocabulary segmentation and detection tasks, achieving knowledge sharing between the two tasks through joint training.
A Simple Recipe to Meta-Learn Forward and Backward Transfer
Edoardo Cetin (King's College London), Oya Celiktutan (King's College London)
Meta LearningImage
🎯 What it does: A minimalistic meta-learning framework SiM4C is proposed to address the forward and backward transfer issues in continual learning, which can be seamlessly integrated into any memory-based continual learning algorithm.
A Simple Vision Transformer for Weakly Semi-supervised 3D Object Detection
Dingyuan Zhang (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A weakly semi-supervised 3D object detection framework is proposed, utilizing a small amount of fully labeled data and a large amount of point-level annotations for training.
A Skeletonization Algorithm for Gradient-Based Optimization
Martin J. Menten (Technical University of Munich), Daniel Rueckert (Imperial College London)
SegmentationOptimizationImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A three-dimensional skeletonization algorithm capable of backpropagating gradient optimization is proposed, which can extract thin skeletons while maintaining the topological structure.
A Soft Nearest-Neighbor Framework for Continual Semi-Supervised Learning
Zhiqi Kang (Universite Grenoble Alpes), Karteek Alahari
ClassificationKnowledge DistillationContrastive LearningImage
🎯 What it does: A soft nearest neighbor framework for continuous semi-supervised learning, NNCSL, is proposed.
A step towards understanding why classification helps regression
Silvia L. Pintea (Delft University of Technology), Jan C. van Gemert (Delft University of Technology)
Depth EstimationConvolutional Neural NetworkImageVideo
🎯 What it does: This paper explores why adding classification loss can improve performance in deep regression tasks, and clarifies that its main role is to compensate for the impact of uneven sample distribution.
A Theory of Topological Derivatives for Inverse Rendering of Geometry
Ishit Mehta (University of California San Diego), Ravi Ramamoorthi (University of California San Diego)
GenerationOptimizationDiffusion modelImageMesh
🎯 What it does: This paper proposes a differentiable surface evolution theory that utilizes topological derivatives to achieve discrete topological changes in inverse rendering, allowing for the generation of holes and phases within shapes, thereby overcoming the limitations of traditional methods that rely solely on shape derivatives.
A Unified Continual Learning Framework with General Parameter-Efficient Tuning
Qiankun Gao (Peking University), Jian Zhang (Peking University)
ClassificationOptimizationTransformerPrompt EngineeringImage
🎯 What it does: A general Learning-Accumulation-Integration (LAE) framework is proposed, which utilizes any parameter-efficient tuning (PET) modules (such as Adapter, LoRA, Prefix) for continual learning on pre-trained models, and achieves memoryless continual learning through the integration of online and offline PET modules.
A Unified Framework for Robustness on Diverse Sampling Errors
Myeongho Jeon (Seoul National University), Joonseok Lee (Seoul National University)
Domain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A unified framework is proposed that utilizes Instance-level Adaptive Inference (IAI) to dynamically adjust the feature space during inference, enhancing the model's robustness under different sampling errors (such as single-domain generalization and unbiased learning).
A-STAR: Test-time Attention Segregation and Retention for Text-to-image Synthesis
Aishwarya Agarwal (Adobe Research), Balaji Vasan Srinivasan (Adobe Research)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper proposes an optimization method for pre-trained text-to-image diffusion models during the inference phase using two types of attention losses (attention separation loss and attention retention loss) to address the issues of concept overlap and information decay.
A2Q: Accumulator-Aware Quantization with Guaranteed Overflow Avoidance
Ian Colbert (Advanced Micro Devices), Jakoba Petri-Koenig (Advanced Micro Devices)
ClassificationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: The A2Q method is proposed, which trains quantized neural networks to use low-precision accumulators during inference without overflow.
Ablating Concepts in Text-to-Image Diffusion Models
Nupur Kumari (Carnegie Mellon University), Jun-Yan Zhu
GenerationData SynthesisPrompt EngineeringDiffusion modelImageText
🎯 What it does: Efficiently eliminate specified concepts (such as artistic styles, specific instances, or memory images) on a pre-trained text-to-image diffusion model without retraining the entire model.
AccFlow: Backward Accumulation for Long-Range Optical Flow
Guangyang Wu (University of Electronic Science and Technology of China), Wenyi Wang (University of Electronic Science and Technology of China)
Optical FlowVideo
🎯 What it does: This paper proposes a recursive framework named AccFlow for estimating long-range optical flow by backward accumulating local optical flow.
Accurate 3D Face Reconstruction with Facial Component Tokens
Tianke Zhang (Tsinghua University), Yu Li
GenerationPose EstimationTransformerImageVideo
🎯 What it does: We propose TokenFace, which uses six independent facial component tokens to regress FLAME parameters in a Transformer, achieving precise 3D facial reconstruction from a single image.
Accurate and Fast Compressed Video Captioning
Yaojie Shen (Institute of Software, Chinese Academy of Sciences), Libo Zhang (Institute of Software, Chinese Academy of Sciences)
GenerationCompressionComputational EfficiencyTransformerVideoText
🎯 What it does: Generate video subtitles directly in the compressed domain (I-frames, motion vectors, and residuals) in an end-to-end manner, eliminating the need to decode video frames or offline extract multimodal features.
Achievement-Based Training Progress Balancing for Multi-Task Learning
Hayoung Yun (Samsung Research), Hanjoo Cho (Samsung Research)
Object DetectionSegmentationDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: A task completion-based multi-task loss function (Achievement-based Multi-Task Loss, AMTL) is proposed, which dynamically adjusts weights through the ratio of task accuracy to single-task benchmarks, and uses a weighted geometric mean to avoid one task dominating the loss.
ACLS: Adaptive and Conditional Label Smoothing for Network Calibration
Hyekang Park (Yonsei University), Bumsub Ham (Korea Institute of Science and Technology)
ClassificationSegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new regularization method during training called ACLS, aimed at improving the probability calibration performance of deep neural networks.
ActFormer: A GAN-based Transformer towards General Action-Conditioned 3D Human Motion Generation
Liang Xu (SenseTime Research), Wei Wu (SenseTime Research)
GenerationData SynthesisTransformerGenerative Adversarial NetworkVideo
🎯 What it does: A GAN-based Transformer framework called ActFormer is proposed for generating 3D human motion for single and multiple person interactions based on semantic action labels.
Action Sensitivity Learning for Temporal Action Localization
Jiayi Shao (Zhejiang University), Yi Yang (Alibaba Group)
RecognitionObject DetectionTransformerContrastive LearningVideo
🎯 What it does: This paper proposes an action-sensitive learning framework (ASL) for video temporal action localization, which can dynamically adjust the training process by learning the contribution of each frame to the classification and localization sub-tasks.
Activate and Reject: Towards Safe Domain Generalization under Category Shift
Chaoqi Chen (University of Hong Kong), Yizhou Yu (University of Hong Kong)
ClassificationObject DetectionSegmentationDomain AdaptationImage
🎯 What it does: The ART (Activate and Reject) framework is proposed to address the issue of category drift under domain generalization, balancing unknown class detection and known class classification.
Active Neural Mapping
Zike Yan (Peking University), Hongbin Zha (Peking University)
Robotic IntelligenceReinforcement LearningSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This study investigates active neural mapping, utilizing implicit neural fields with continual learning to actively explore unknown environments and construct high-precision 3D scene maps in real-time.
Active Self-Supervised Learning: A Few Low-Cost Relationships Are All You Need
Vivien Cabannes (Meta AI), Randall Balestriero (Meta AI)
Representation LearningContrastive LearningImage
🎯 What it does: The paper proposes an active self-supervised learning framework called PAL, which efficiently constructs similarity graphs and learns representations without the need for complete labels through a small number of low-cost similarity relationship queries.
Active Stereo Without Pattern Projector
Luca Bartolomei (University of Bologna), Stefano Mattoccia (University of Bologna)
Depth EstimationImage
🎯 What it does: A virtual pattern projection (VPP) framework is proposed, which synthesizes 'virtual' projection images using sparse depth sensors in scenes without projectors, significantly improving the accuracy of stereo matching.
ACTIVE: Towards Highly Transferable 3D Physical Camouflage for Universal and Robust Vehicle Evasion
Naufal Suryanto (Pusan National University), Howon Kim (Pusan National University)
Object DetectionAutonomous DrivingAdversarial AttackImage
🎯 What it does: A physical attack framework for three-dimensional vehicles, ACTIVE, has been designed to achieve generalized and robust full coverage camouflage.
ActorsNeRF: Animatable Few-shot Human Rendering with Generalizable NeRFs
Jiteng Mu (University of California San Diego), Xiaolong Wang (University of California San Diego)
GenerationData SynthesisNeural Radiance FieldVideo
🎯 What it does: Developed ActorsNeRF, a category-level generalizable NeRF model that can generate high-quality, animatable multi-view renderings of new actors using only a few monocular video frames.
Ada3D : Exploiting the Spatial Redundancy with Adaptive Inference for Efficient 3D Object Detection
Tianchen Zhao (Tsinghua University), Yu Wang (Tsinghua University)
Object DetectionAutonomous DrivingComputational EfficiencyPoint Cloud
🎯 What it does: We propose Ada3D, an adaptive reasoning framework that significantly reduces the computational and memory costs of 3D object detection by filtering out redundant spatial information in 3D voxels and BEV features.
AdaMV-MoE: Adaptive Multi-Task Vision Mixture-of-Experts
Tianlong Chen (University of Texas at Austin), Yeqing Li (Google)
ClassificationObject DetectionSegmentationTransformerMixture of ExpertsImage
🎯 What it does: This paper proposes AdaMV-MoE, an adaptive multi-task visual mixture expert framework designed to simultaneously address tasks such as image classification, object detection, and instance segmentation.
AdaNIC: Towards Practical Neural Image Compression via Dynamic Transform Routing
Lvfang Tao (Peking University), Chenhao Zhang (Peking University)
CompressionKnowledge DistillationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: The AdaNIC framework is proposed, which implements a variable-width compressed autoencoder (SA-CAE) through dynamic capacity routing, and combines the JUMC decision threshold with a lightweight routing agent to achieve content-aware computation allocation.
ADAPT: Efficient Multi-Agent Trajectory Prediction with Adaptation
Görkay Aydemir (Koc University), Fatma Güney (Koc University)
Autonomous DrivingComputational EfficiencyTransformerAgentic AITime Series
🎯 What it does: An efficient multi-agent trajectory prediction framework named ADAPT is proposed, capable of unifying trajectory predictions in both single-agent and multi-agent scenarios.
Adaptive and Background-Aware Vision Transformer for Real-Time UAV Tracking
Shuiwang Li (Guilin University of Technology), Xucheng Wang (Guilin University of Technology)
Object TrackingTransformerVideo
🎯 What it does: A real-time UAV tracking framework called Aba-ViTrack is proposed, which utilizes an efficient Vision Transformer (ViT) and a background-aware adaptive token skipping mechanism to achieve feature learning and template-search coupling within a single network.
Adaptive Calibrator Ensemble: Navigating Test Set Difficulty in Out-of-Distribution Scenarios
Yuli Zou (Hong Kong Polytechnic University), Liang Zheng (Australian National University)
Domain AdaptationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper addresses the issue of model calibration failure on out-of-distribution (OOD) data by proposing a method based on Adaptive Calibrator Ensemble (ACE);
Adaptive Frequency Filters As Efficient Global Token Mixers
Zhipeng Huang (University of Science and Technology of China), Baining Guo (Microsoft Research Asia)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: Proposes Adaptive Frequency Filtering (AFF) as an efficient global token mixer and constructs a lightweight visual network called AFFNet based on it.
Adaptive Illumination Mapping for Shadow Detection in Raw Images
Jiayu Sun (Dalian University of Technology), Rynson Lau (City University of Hong Kong)
Object DetectionSegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Using raw RAW images instead of traditional sRGB images for shadow detection, an Adaptive Illumination Mapping (AIM) module is proposed to generate sRGB images with different intensity ranges, and a feedback mechanism is employed to guide AIM in producing images with more shadow contrast, thereby improving shadow detection accuracy.
Adaptive Image Anonymization in the Context of Image Classification with Neural Networks
Nadiya Shvai (Cyclope.ai), Amir Nakib (University Paris Est Creteil)
ClassificationOptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: A method has been developed that utilizes gradient descent to fine-tune the anonymized areas of images, allowing the original classifier to maintain the same category predictions as the original images after anonymization.
Adaptive Nonlinear Latent Transformation for Conditional Face Editing
Zhizhong Huang (Fudan University), Hongming Shan (Fudan University)
Image TranslationGenerationFlow-based ModelGenerative Adversarial NetworkImage
🎯 What it does: Conditional editing of the latent space of StyleGAN is achieved through gradual nonlinear transformations for controllable modification of facial attributes.
Adaptive Positional Encoding for Bundle-Adjusting Neural Radiance Fields
Zelin Gao (Zhejiang University), Yu Zhang (Zhejiang University)
OptimizationNeural Radiance FieldOptical FlowImage
🎯 What it does: Proposes Adaptive Position Encoding (APE) and Periodic Activation Multi-Layer Perceptron (PMLP) to jointly train neural radiance fields, achieving scene reconstruction and camera calibration from unknown camera poses (even intrinsic parameters).
Adaptive Reordering Sampler with Neurally Guided MAGSAC
Tong Wei (Czech Technical University in Prague), Daniel Barath (ETH Zurich)
OptimizationRobotic IntelligenceSpiking Neural NetworkReinforcement LearningImage
🎯 What it does: An adaptive reordering sampler and a neural-guided MAGSAC algorithm are proposed for robust two-view geometric estimation.
Adaptive Rotated Convolution for Rotated Object Detection
Yifan Pu (Tsinghua University), Gao Huang (Tsinghua University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: An Adaptive Rotational Convolution (ARC) module is proposed, which can dynamically rotate convolution kernels according to the orientation of targets in different images, and enhance the feature representation capability of the backbone by combining multiple rotated kernels through a conditional computation mechanism.
Adaptive Similarity Bootstrapping for Self-Distillation Based Representation Learning
Tim Lebailly (KU Leuven), Tinne Tuytelaars (KU Leuven)
Knowledge DistillationRepresentation LearningContrastive LearningImage
🎯 What it does: This paper focuses on the self-distillation framework in self-supervised representation learning, investigating the feasibility of using nearest neighbors (NN) to guide positive sample pairs, and proposes an adaptive similarity guidance method (AdaSim).
Adaptive Spiral Layers for Efficient 3D Representation Learning on Meshes
Francesca Babiloni (Huawei), Stefanos Zafeiriou (Imperial College London)
Computational EfficiencyRepresentation LearningGraph Neural NetworkMesh
🎯 What it does: This paper proposes an adaptive spiral convolution layer suitable for 3D meshes, which can dynamically adjust the length and weights of the spiral path according to the mesh structure, thereby achieving efficient feature learning with a global receptive field and local refinement.
Adaptive Superpixel for Active Learning in Semantic Segmentation
Hoyoung Kim (POSTECH), Jungseul Ok (POSTECH)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: An adaptive superpixel active learning framework is designed for semantic segmentation.
Adaptive Template Transformer for Mitochondria Segmentation in Electron Microscopy Images
Yuwen Pan (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
SegmentationTransformerImage
🎯 What it does: An adaptive template transformer (ATFormer) model is proposed for the automatic segmentation of mitochondria in electron microscope images.
Adaptive Testing of Computer Vision Models
Irena Gao (Stanford University), Marco Tulio Ribeiro (Microsoft Research)
ClassificationObject DetectionRetrievalTransformerLarge Language ModelImageTextRetrieval-Augmented Generation
🎯 What it does: This paper presents AdaVision, a visual model testing process for human-computer interaction that helps users identify and fix semantically consistent failure modes of models.
Adding Conditional Control to Text-to-Image Diffusion Models
Lvmin Zhang (Stanford University), Maneesh Agrawala (Stanford University)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Design and implement the ControlNet architecture, allowing users to finely control the spatial layout of generated images through additional conditional images (such as edge maps, pose skeletons, depth maps, etc.) without compromising the quality of the original large-scale pre-trained diffusion models (like Stable Diffusion).
ADNet: Lane Shape Prediction via Anchor Decomposition
Lingyu Xiao (Southeast University), Wankou Yang (Southeast University)
Object DetectionAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: This paper proposes ADNet, which achieves variable starting point lane line detection by decomposing anchor points through starting point heatmaps and direction prediction, and improves accuracy by combining large kernel attention and GLIoU loss.
Advancing Example Exploitation Can Alleviate Critical Challenges in Adversarial Training
Yao Ge (Nanjing University of Posts and Telecommunications), Xianzhong Long (Nanjing University of Posts and Telecommunications)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper studies example exploitation in adversarial training, first proposing a robust confidence metric that divides samples into accuracy-critical (A-C) and robustness-critical (R-C) categories, and analyzes their different contributions to model accuracy and robustness. Subsequently, a new example handling method is designed, which reduces the robustness learning intensity for A-C samples and enhances the robustness learning intensity for R-C samples (through adaptive λ or step size). This method is applied to both multi-step (TRADES, TEAT, etc.) and single-step (FastAT, GradAlign, etc.) adversarial training. Experiments show that it can simultaneously alleviate the accuracy-robustness trade-off, robustness overfitting, and catastrophic overfitting issues.
Advancing Referring Expression Segmentation Beyond Single Image
Yixuan Wu (Zhejiang University), Rui Zhao (SenseTime Research)
Object DetectionSegmentationTransformerImageBenchmark
🎯 What it does: This paper proposes a Generalized Representation Segmentation (GRES) task for multi-image collections and constructs the corresponding GRD dataset and benchmark model GRSer.
AdvDiffuser: Natural Adversarial Example Synthesis with Diffusion Models
Xinquan Chen (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), Cheng-Zhong Xu (University of Macau)
GenerationData SynthesisAdversarial AttackDiffusion modelImage
🎯 What it does: A natural unconstrained adversarial example (UAE) generation method based on diffusion models, AdvDiffuser, is proposed, which can generate natural and highly deceptive adversarial images from scratch or conditioned on reference images.
AdVerb: Visually Guided Audio Dereverberation
Sanjoy Chowdhury (University of Maryland), Dinesh Manocha (University of Maryland)
RecognitionRestorationTransformerImageMultimodalityAudio
🎯 What it does: A visual information-based audio dereverberation framework called AdVerb is proposed, which jointly estimates clear speech using scene images and reverberant audio.
Adversarial Bayesian Augmentation for Single-Source Domain Generalization
Sheng Cheng (Arizona State University), Yezhou Yang (University of Maryland Baltimore County)
Data SynthesisDomain AdaptationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a new single-source domain generalization method called Adversarial Bayesian Augmentation (ABA), which enhances model performance in unknown domains by introducing Bayesian neural networks in convolutional layers and combining them with adversarial training to generate diverse image augmentations.
Adversarial Finetuning with Latent Representation Constraint to Mitigate Accuracy-Robustness Tradeoff
Satoshi Suzuki (NTT Computer and Data Science Laboratories), Ryo Masumura (NTT Computer and Data Science Laboratories)
Knowledge DistillationRepresentation LearningAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A new adversarial training method called ARREST is proposed, which combines adversarial fine-tuning, representation-guided knowledge distillation, and noise replay to alleviate the trade-off between standard accuracy and robustness.
Adverse Weather Removal with Codebook Priors
Tian Ye (National University of Singapore), Yun Liu (Southwest University)
RestorationTransformerGenerative Adversarial NetworkImage
🎯 What it does: A unified adverse weather removal network AWRCP based on VQGAN codebook prior is designed and implemented, capable of simultaneously removing various degradations such as fog, rain, snow, and raindrops in a single forward pass.
AerialVLN: Vision-and-Language Navigation for UAVs
Shubo Liu (Northwestern Polytechnical University), Qi Wu (University of Adelaide)
Robotic IntelligenceTransformerReinforcement LearningVision Language ModelTextMultimodality
🎯 What it does: AerialVLN, a visual and language navigation task for urban-level drones, is proposed, along with the corresponding simulator and large-scale dataset.
AesPA-Net: Aesthetic Pattern-Aware Style Transfer Networks
Kibeom Hong (Yonsei University), Hyeran Byun (Yonsei University)
Image TranslationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: AesPA-Net is proposed, utilizing pattern repeatability metrics to adaptively fuse attention-based and global statistical style transfer for artistic style conversion.
Affective Image Filter: Reflecting Emotions from Text to Images
Shuchen Weng (Peking University), Boxin Shi (Peking University)
Image TranslationGenerationData SynthesisTransformerVision Language ModelGenerative Adversarial NetworkImageTextMultimodality
🎯 What it does: This paper proposes the Emotion Image Filter (AIF) task, which utilizes a multimodal Transformer to convert abstract emotions in text into visually concrete images that align with emotional semantics.
Affine-Consistent Transformer for Multi-Class Cell Nuclei Detection
Junjia Huang (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
Object DetectionSegmentationTransformerImage
🎯 What it does: An Affine-Consistent Transformer (AC-Former) is proposed, which directly outputs the positions and categories of multi-class cell nuclei through collaborative training of global-local networks, without the need for post-processing.
AffordPose: A Large-Scale Dataset of Hand-Object Interactions with Affordance-Driven Hand Pose
Juntao Jian (Dalian University of Technology), Jian Liu (Tsinghua University)
SegmentationGenerationPose EstimationConvolutional Neural NetworkGraph Neural NetworkGenerative Adversarial NetworkPoint CloudMesh
🎯 What it does: This paper proposes and publicly releases the AffordPose dataset, which records 26.7K hand-object interaction samples. Each sample includes 3D objects, part-level affordance annotations, and hand pose parameters, and based on this, experiments on hand-object affordance understanding and affordance-driven interaction generation are conducted.
AG3D: Learning to Generate 3D Avatars from 2D Image Collections
Zijian Dong (ETH Zurich), Andreas Geiger (University of Tuebingen)
GenerationData SynthesisPose EstimationGenerative Adversarial NetworkImageVideo
🎯 What it does: Learn to generate animatable 3D human avatars from unstructured 2D image collections, with control over pose, identity, and clothing.
AGG-Net: Attention Guided Gated-Convolutional Network for Depth Image Completion
Dongyue Chen (Northeastern University), Tong Jia (Northeastern University)
RestorationDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: For RGB-D image depth completion, a novel Attention Guided Gated Convolution Network (AGG-Net) is proposed to achieve fine recovery of missing depth.
Agglomerative Transformer for Human-Object Interaction Detection
Danyang Tu (Shanghai Jiao Tong University), Wei Shen (Shanghai Jiao Tong University)
RecognitionObject DetectionTransformerImage
🎯 What it does: A single-stage Transformer-based framework for person-object interaction detection, AGER, is proposed, which generates complete instance tokens internally in the encoder using text-guided dynamic clustering, thereby extracting complete instance-level features without the need for an additional object detector or instance decoder.