ECCV 2024 Papers with AI Summaries
European Conference on Computer Vision · 2387 papers
→ ECCV 2024 papers with code (980)
Each paper below shows an AI-generated one-line summary. Get the full 6-part summary (innovation, method, data, results, limitations) and search all 2387 ECCV 2024 papers by keyword, author or institution —
free trial on arXivSub.
"A Framework for Efficient Model Evaluation through Stratification, Sampling, and Estimation"
Riccardo Fogliato (Amazon Web Services), Pietro Perona (Amazon Web Services)
Computational EfficiencyData-Centric LearningImageBenchmark
🎯 What it does: Propose a statistical framework combining stratification, sampling, and estimation to achieve high-precision estimation of model performance on test sets with only a small number of labeled samples.
"BK-SDM: A Lightweight, Fast, and Cheap Version of Stable Diffusion"
Bo-Kyeong Kim (Nota Inc), Shinkook Choi (Nota Inc)
GenerationComputational EfficiencyKnowledge DistillationSupervised Fine-TuningDiffusion modelImage
🎯 What it does: Compress the U-Net structure of Stable Diffusion using block pruning and feature distillation to build a lightweight model, supporting fast text generation, personalized generation, and image-to-image conversion.
"BlinkVision: A Benchmark for Optical Flow, Scene Flow and Point Tracking Estimation using RGB Frames and Events"
Yijin Li (State Key Lab of CAD&CG, Zhejiang University), Hongsheng Li (CUHK MMLab)
Data SynthesisVision Language ModelOptical FlowMultimodalityBenchmark
🎯 What it does: This paper constructs the BlinkVision benchmark, providing synchronized RGB frames and event data, generating dense annotations for three corresponding tasks: optical flow, scene flow, and point tracking, covering 410 daily categories, with highly naturalized and diverse scenarios.
"ByteEdit: Boost, Comply and Accelerate Generative Image Editing"
Yuxi Ren (ByteDance), Lean FU
GenerationTransformerReinforcement LearningVision Language ModelDiffusion modelGenerative Adversarial NetworkImageTextMultimodality
🎯 What it does: Proposed the ByteEdit framework, which enhances the quality, coherence, text adherence, and generation speed of generative image editing through feedback learning.
"Category-level Object Detection, Pose Estimation and Reconstruction from Stereo Images"
Chuanrui Zhang (Tsinghua University), Haoqian Wang (Tsinghua University)
Object DetectionPose EstimationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: Proposed a single-stage end-to-end CODERS framework that simultaneously achieves category-level object detection, 6D pose estimation, and 3D shape reconstruction using stereo images.
"Clearer Frames, Anytime: Resolving Velocity Ambiguity in Video Frame Interpolation"
Zhihang Zhong (Shanghai Artificial Intelligence Laboratory), Jian Wang (Snap Inc)
RestorationOptical FlowVideo
🎯 What it does: This paper addresses the speed ambiguity problem in video frame interpolation by proposing two interpolation strategies: distance indexing and iterative reference-based estimation, which are integrated as plug-in methods into existing arbitrary time interpolation models.
"Close, But Not There: Boosting Geographic Distance Sensitivity in Visual Place Recognition"
Sergio Izquierdo (University of Zaragoza), Javier Civera (University of Zaragoza)
RetrievalTransformerContrastive LearningImageGraph
🎯 What it does: Proposed a new sampling strategy called CliqueMining to construct challenging batches in visual place recognition (VPR) training, thereby enhancing geospatial distance sensitivity (GDS).
"DECOLLAGE: 3D Detailization by Controllable, Localized, and Learned Geometry Enhancement"
Qimin Chen (Simon Fraser University), Siddhartha Chaudhuri (Adobe Research)
GenerationGenerative Adversarial Network
🎯 What it does: Proposes a controllable local 3D detailing method based on Pyramid GAN, allowing users to paint style regions on coarse voxel models, with the network generating high-resolution geometric details according to the specified styles.
"EMDM: Efficient Motion Diffusion Model for Fast, High-Quality Human Motion Generation"
Wenyang Zhou (University of Cambridge), Lingjie Liu (University of Pennsylvania)
GenerationData SynthesisTransformerDiffusion modelGenerative Adversarial NetworkMultimodalitySequential
🎯 What it does: Proposed an efficient motion diffusion model (EMDM) that can generate high-quality, diverse human motion with very few sampling steps.
"Eyes Closed, Safety On: Protecting Multimodal LLMs via Image-to-Text Transformation"
Yunhao Gou (Southern University of Science and Technology), Yu Zhang (Huawei Noah's Ark Lab)
Safty and PrivacyLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Proposes a training-free protection framework called ECSO, which enhances the security of multi-modal large language models (MLLM) against malicious image inputs through self-detection and image-to-text conversion.
"FARSE-CNN: Fully Asynchronous, Recurrent and Sparse Event-Based CNN"
Riccardo Santambrogio (Politecnico di Milano), Matteo Matteucci (Politecnico di Milano)
ClassificationRecognitionObject DetectionConvolutional Neural NetworkRecurrent Neural Network
🎯 What it does: This paper proposes a fully asynchronous, recursive, and sparse event convolutional network called FARSE-CNN, designed to efficiently process the sparse spatiotemporal streams generated by event cameras.
"Hyperion – A fast, versatile symbolic Gaussian Belief Propagation framework for Continuous-Time SLAM"
David Hug (ETH Zürich), Margarita Chli (ETH Zürich)
OptimizationRobotic IntelligenceSimultaneous Localization and MappingImage
🎯 What it does: Proposed the Hyperion framework, combining fast continuous-time B-/Z-spline parameterization and efficient Gaussian Bayes propagation (GBP) algorithm, achieving real-time inference for distributed continuous-time SLAM;
"Idling Neurons, Appropriately Lenient Workload During Fine-tuning Leads to Better Generalization"
Hongjing Niu (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
Knowledge DistillationRepresentation LearningSupervised Fine-TuningImage
🎯 What it does: Studied reducing distortion of pre-trained features during fine-tuning through 'idle neurons' or 'relaxing workload' to enhance model generalization on downstream tasks.
"MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training"
Brandon McKinzie (Apple), Yinfei Yang (Apple)
TransformerLarge Language ModelMixture of ExpertsVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper systematically studies and evaluates pre-training methods for multi-modal large language models (MLLMs), explores the impact of image encoders, vision-language connectors, and data mixing on performance, and builds the MM1 series of models based on these insights.
"NeuSDFusion: A Spatial-Aware Generative Model for 3D Shape Completion, Reconstruction, and Generation"
Ruikai Cui (Australian National University), Pan Ji (Australian National University)
GenerationTransformerDiffusion modelNeural Radiance FieldAuto EncoderImageTextMultimodalityPoint CloudMesh
🎯 What it does: Proposes NeuSDFusion, a spatial-aware 3D shape generation framework that utilizes three orthogonal planes combined with continuous SDF representation to achieve high-quality, multi-modal 3D shape generation, completion, reconstruction, and text-driven synthesis.
"Not Just Change the Labels, Learn the Features: Watermarking Deep Neural Networks with Multi-View Data"
Yuxuan Li (Harbin Institute of Technology), Yunhui Guo (University of Texas at Dallas)
Safty and PrivacyAdversarial AttackImage
🎯 What it does: Propose a deep learning model watermarking method called MAT, which utilizes multi-view data to construct a trigger set, thereby enhancing defense against model extraction attacks.
"On Calibration of Object Detectors: Pitfalls, Evaluation and Baselines"
Selim Kuzucu (Five AI Ltd.), Puneet Dokania
Object DetectionImage
🎯 What it does: This paper analyzes the limitations of existing calibration evaluation methods for object detectors and proposes a more reasonable joint evaluation framework, verifying the effectiveness of post-calibration methods.
"Plan, Posture and Go: Towards Open-vocabulary Text-to-Motion Generation"
Jinpeng Liu (Shenzhen Key Laboratory of Ubiquitous Data Enabling, Shenzhen International Graduate School, Tsinghua University), Xin Tong (Microsoft Research Asia)
GenerationTransformerLarge Language ModelVision-Language-Action ModelDiffusion modelMultimodality
🎯 What it does: This paper proposes the PRO-Motion framework, decomposing text-to-motion generation into three steps: Planning (Motion Planner), Posture (Posture-Denoiser), and Movement (Go-Denoiser). First, a large language model (GPT-3.5) generates a series of concise posture scripts based on natural language prompts. Subsequently, a diffusion model maps the scripts to key postures, and the Viterbi planner selects the most reasonable posture sequence. Finally, another diffusion model interpolates the key postures and predicts full-body translation and rotation, resulting in complete 3D motion.
"PointNeRF++: A multi-scale, point-based Neural Radiance Field"
Weiwei Sun (University of British Columbia), Kwang Moo Yi (University of British Columbia)
GenerationNeural Radiance FieldGaussian SplattingPoint Cloud
🎯 What it does: What they did: Proposed PointNeRF++, enhancing NeRF rendering for sparse or incomplete point clouds through multi-scale voxel aggregation, global voxels, and tri-plane features.
"PoseEmbroider: Towards a 3D, Visual, Semantic-aware Human Pose Representation"
Ginger Delmas, Gregory Rogez
Pose EstimationGraph Neural NetworkImage
🎯 What it does: This paper proposes a novel human pose estimation method based on graph convolutional networks (GCN), utilizing the graph structure between joints for feature learning;
"ProCreate, Don't Reproduce! Propulsive Energy Diffusion for Creative Generation"
Jack Lu (New York University), Mengye Ren (New York University)
GenerationDiffusion modelScore-based ModelImage
🎯 What it does: Proposes the ProCreate method, which introduces energy guidance during the diffusion model inference process to make generated image embeddings diverge from reference images, thereby enhancing sample diversity and creativity while preventing training data duplication.
"Propose, Assess, Search: Harnessing LLMs for Goal-Oriented Planning in Instructional Videos"
Md Mohaiminul Islam, Xitong Yang (Meta)
TransformerLarge Language ModelVision Language ModelVideoText
🎯 What it does: Proposes the VidAssist framework, which leverages large language models (LLMs) for goal-oriented planning and combines a propose-evaluate-search strategy to achieve zero/few-shot planning.
"Refine, Discriminate and Align: Stealing Encoders via Sample-Wise Prototypes and Multi-Relational Extraction"
Shuchi Wu (NJUST), Tao Xiang (CQU)
Safty and PrivacyRepresentation LearningAdversarial AttackContrastive LearningImageMultimodality
🎯 What it does: The paper proposes the RDA method, which efficiently steals pre-trained encoders under low query costs by using sample-level prototypes and multi-relationship extraction loss.
"RICA^2: Rubric-Informed, Calibrated Assessment of Actions"
Abrar Majeedi (University of Wisconsin-Madison), Yin Li (University of Wisconsin-Madison)
RecognitionGraph Neural NetworkTransformerVision-Language-Action ModelVideo
🎯 What it does: This paper proposes RICA 2, a deep probabilistic model that integrates scoring criteria and uncertainty modeling for video action quality assessment.
"SEA-RAFT: Simple, Efficient, Accurate RAFT for Optical Flow"
Yihan Wang (Princeton University), Jia Deng (Princeton University)
Convolutional Neural NetworkOptical FlowImageVideo
🎯 What it does: This paper proposes SEA-RAFT, a simplified and more efficient and accurate optical flow estimation framework based on RAFT.
"Segment, Lift and Fit: Automatic 3D Shape Labeling from 2D Prompts"
Jianhao Li (Beihang University), Ping Luo (University of Hong Kong)
Object DetectionPose EstimationAutonomous DrivingTransformerNeural Radiance FieldImagePoint Cloud
🎯 What it does: Propose a 3D automatic annotation method called SLF based on 2D point or bounding box prompts, which can generate 3D shapes and poses of objects from images.
"SIMBA: Split Inference - Mechanisms, Benchmarks and Attacks"
Abhishek Singh (Massachusetts Institute of Technology), Ramesh Raskar (Massachusetts Institute of Technology)
Safty and PrivacyComputational EfficiencyConvolutional Neural NetworkGenerative Adversarial NetworkImageBenchmark
🎯 What it does: Built an expandable benchmark platform to evaluate representation obfuscation techniques and reconstruction attacks in image data privacy protection.
"SLAck: Semantic, Location, and Appearance Aware Open-Vocabulary Tracking"
Siyuan Li (Eth Zurich), Luc Van Gool (Eth Zurich)
Object TrackingConvolutional Neural NetworkGraph Neural NetworkTransformerVision Language ModelContrastive LearningVideoMultimodality
🎯 What it does: Proposed the SLAck framework, achieving unified fusion of semantic, location, and appearance features in open-vocabulary multi-object tracking (MOT), and implementing end-to-end association through a lightweight spatiotemporal object graph.
"Smoothness, Synthesis, and Sampling: Re-thinking Unsupervised Multi-View Stereo with DIV Loss"
Alex Rich (University of California, Santa Barbara), Tobias Hollerer
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new unsupervised multi-view stereo (MVS) loss function called DIV loss, primarily through three improvements: 1) Using second-order gradient clipping for depth smoothing constraints to better handle object edges; 2) Utilizing a small CNN to learn weights for weighted synthesis of multi-view images, improving perspective-dependent image synthesis loss; 3) Using additional views beyond input perspectives during supervision to enhance model generalization and robustness.
"SPHINX: A Mixer of Weights, Visual Embeddings and Image Scales for Multi-modal Large Language Models"
Ziyi Lin (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Designed and trained the SPHINX multimodal large language model, integrating weighted mixing, visual embedding mixing, and multiscale high-resolution image processing to enhance multimodal understanding and fine-grained visual reasoning capabilities.
"Towards Dual Transparent Liquid Level Estimation in Biomedical Lab: Dataset, Methods and Practice"
Xiayu Wang (Huazhong University of Science and Technology), Tian Xia (Huazhong University of Science and Technology)
Object DetectionPose EstimationConvolutional Neural NetworkImageBiomedical DataBenchmark
🎯 What it does: Aiming at the liquid surface estimation task for dual transparent liquids (both liquid and container are transparent) in AI-guided bio-laboratories, we propose the DTLD dataset and an end-to-end TCLD method for detecting liquid contact lines and estimating liquid surface height.
"Tracking Meets LoRA: Faster Training, Larger Model, Stronger Performance"
Liting Lin (Pengcheng Laboratory), Haibin Ling (Stony Brook University)
Object TrackingTransformerVideo
🎯 What it does: In the visual object tracking task, the parameter-efficient fine-tuning technique LoRA is applied to a Transformer-based single-stream tracking framework, and a decoupled position embedding and token type embedding are introduced, along with a head network that uses only MLP, constructing a series of large-scale tracking models (LoRAT) that can be trained under low memory consumption.
"UniINR: Event-guided Unified Rolling Shutter Correction, Deblurring, and Interpolation"
Yunfan Lu (Hong Kong University of Science and Technology), Hui Xiong (Hong Kong University of Science and Technology)
RestorationNeural Radiance FieldVideo
🎯 What it does: Propose the UniINR framework, which utilizes event camera data to simultaneously perform rolling shutter (RS) correction, deblurring, and high-frame-rate global shutter (GS) frame interpolation based on a single RS blurry frame, recovering clear GS frames at arbitrary time points.
"Unsupervised, Online and On-The-Fly Anomaly Detection For Non-Stationary Image Distributions"
Declan GD McIntosh, Alexandra Branzan Albu (University of Victoria)
Anomaly DetectionConvolutional Neural NetworkImageBenchmark
🎯 What it does: Proposes Online-InReaCh, an algorithm capable of detecting and locating image anomalies online in real-time in an unsupervised manner.
"Veil Privacy on Visual Data: Concealing Privacy for Humans, Unveiling for DNNs"
Shuchao Pang (Nanjing University of Science and Technology), Yazhou Yao (Nanjing University of Science and Technology)
Safty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: Propose the Veil Privacy framework, which generates concealed data invisible to the human eye but usable for DNN training and inference through random pixel flipping and gradient iterative algorithms, achieving a balance between visual data privacy and model usability.
"X-InstructBLIP: A Framework for Aligning Image, 3D, Audio, Video to LLMs and its Emergent Cross-modal Reasoning"
Artemis Panagopoulou (University of Pennsylvania), Juan Carlos Niebles (Salesforce AI Research)
Representation LearningTransformerLarge Language ModelVision Language ModelContrastive LearningImageVideoMultimodalityPoint CloudAudio
🎯 What it does: Proposed the X-InstructBLIP framework, aligning multi-modal data such as images, 3D, audio, and videos with a frozen large language model (LLM), and achieved cross-modal reasoning capabilities.
∞-Brush: Controllable Large Image Synthesis with Diffusion Models in Infinite Dimensions
Minh-Quan Le (Stony Brook University), Dimitris Samaras (Stony Brook University)
GenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelContrastive LearningImageBiomedical Data
🎯 What it does: Proposed a controllable large-image generation model ∞-Brush in an infinite-dimensional function space, achieving effective injection of conditional information through cross-attention neural operators.
2S-ODIS: Two-Stage Omni-Directional Image Synthesis by Geometric Distortion Correction
Atsuya Nakata (Sophia University), Takao Yamanaka (Sophia University)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderImage
🎯 What it does: Propose a two-stage panoramic image synthesis method called 2S-ODIS, which generates high-quality panoramic images using a pre-trained VQGAN while significantly reducing training time.
3D Congealing: 3D-Aware Image Alignment in the Wild
Yunzhi Zhang (Stanford University), Varun Jampani (Stability AI)
Pose EstimationDiffusion modelScore-based ModelNeural Radiance FieldImage
🎯 What it does: Propose the 3D Congealing task, aligning images of different instances in a shared 3D space and supporting downstream applications such as pose estimation and image editing.
3D Gaussian Parametric Head Model
Yuelang Xu (Tsinghua University), Yebin Liu (Tsinghua University)
GenerationGaussian SplattingImagePoint Cloud
🎯 What it does: Propose a 3D Gaussian Parametric Head Model, achieving high-fidelity, real-time rendering of head avatars, supporting single-image fitting, expression, and identity editing.
3D Hand Pose Estimation in Everyday Egocentric Images
Aditya Prakash (University of Illinois Urbana-Champaign), Saurabh Gupta (University of Illinois Urbana-Champaign)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes the WildHands system for predicting complete 3D hand poses from single RGB images in everyday first-person perspective images.
3D Hand Sequence Recovery from Real Blurry Images and Event Stream
JoonKyu Park, Kyoung Mu Lee (Seoul National University)
Pose EstimationTransformerImageSequential
🎯 What it does: Proposes a method called EBHNet for recovering 3D hand sequences from real motion-blurred hand images and event streams, supporting inputs of single images, single events, or a combination of both;
3D Human Pose Estimation via Non-Causal Retentive Networks
Kaili Zheng (Tsinghua University), Ji Wu (Tsinghua University)
Pose EstimationRecurrent Neural NetworkVideo
🎯 What it does: Propose a 3D human pose estimation model NC-RetNet based on RetNet, which performs inference by utilizing a large number of past frames and a small number of future frames.
3D Open-Vocabulary Panoptic Segmentation with 2D-3D Vision-Language Distillation
Zihao Xiao (Johns Hopkins University), Shiwei Sheng (Google DeepMind)
SegmentationAutonomous DrivingKnowledge DistillationTransformerPoint Cloud
🎯 What it does: Proposed the first 3D open-vocabulary domain panoptic segmentation method, integrating LiDAR-learned features with frozen CLIP visual features, and using a unified classification head to predict base classes and novel classes;
3D Reconstruction of Objects in Hands without Real World 3D Supervision
Aditya Prakash (University of Illinois Urbana Champaign), Saurabh Gupta (University of Illinois Urbana Champaign)
GenerationPose EstimationNeural Radiance FieldGenerative Adversarial NetworkImageVideo
🎯 What it does: Reconstruct the 3D shape of handheld objects from a single RGB image, without relying on real-world 3D supervision.
3D Single-object Tracking in Point Clouds with High Temporal Variation
Qiao Wu (Northwestern Polytechnical University), Jiaqi Yang (Northwestern Polytechnical University)
Object TrackingAutonomous DrivingGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: Proposes a 3D single-object tracking framework named HVTrack for point clouds with high temporal variations, addressing challenges such as strong deformation, interference from similar targets, and background noise;
3D Small Object Detection with Dynamic Spatial Pruning
Zhihao Sun (University of Science and Technology Beijing), Xiuwei Xu (Tsinghua University)
Object DetectionConvolutional Neural NetworkPoint Cloud
🎯 What it does: Propose a 3D small object detection method called DSPDet3D based on dynamic spatial cropping, significantly reducing the decoder computational cost while maintaining high-resolution features.
3D Weakly Supervised Semantic Segmentation with 2D Vision-Language Guidance
Xiaoxu Xu, Xu Wang (Shenzhen University)
SegmentationConvolutional Neural NetworkVision Language ModelPoint Cloud
🎯 What it does: Propose 3DSS-VLG, a weakly supervised 3D semantic segmentation method that uses a 2D vision-language model as a bridge to achieve implicit alignment between point clouds and text labels.
3D-GOI: 3D GAN Omni-Inversion for Multifaceted and Multi-object Editing
Haoran Li (University of Science and Technology of China), Peng Yuan Zhou (Aarhus University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Proposed a 3D-GOI framework to enable multi-dimensional (scaling, translation, rotation, appearance, shape) editing of images containing multiple objects, and simultaneously edit multiple objects and the background in the same scene.
3DEgo: 3D Editing on the Go!
Umar Khalid (University of Central Florida), Chen Chen (University of Central Florida)
GenerationData SynthesisTransformerPrompt EngineeringDiffusion modelGaussian SplattingVideoTextPoint Cloud
🎯 What it does: Achieved a one-stage 3D editing pipeline by performing 2D autoregressive editing using text prompts in monocular videos and directly constructing 3D Gaussian scenes, avoiding COLMAP and SfM.
3DFG-PIFu: 3D Feature Grids for Human Digitization from Sparse Views
Kennard Yanting Chan (Nanyang Technological University), Weisi Lin (Nanyang Technological University)
GenerationImageMesh
🎯 What it does: Propose a pixel-aligned implicit model called 3DFG-PIFu, which utilizes 3D feature grids to globally fuse multi-view information, refines the human mesh through an iterative mechanism in the backend, and supports an SDF-form SMPL-X prior.
3DGazeNet: Generalizing Gaze Estimation with Weak Supervision from Synthetic Views
Evangelos Ververas (Imperial College London), Stefanos Zafeiriou (Imperial College London)
RecognitionConvolutional Neural NetworkGenerative Adversarial NetworkMesh
🎯 What it does: This paper proposes a general 3D gaze estimation model, 3DGazeNet, based on dense 3D eye mesh regression, achieving cross-domain generalization through weakly supervised synthetic views.
3DSA:Multi-View 3D Human Pose Estimation With 3D Space Attention Mechanisms
Po Han Chen (National Cheng Kung University), Chia-Chi Tsai (National Cheng Kung University)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: Designed and implemented a 3D Spatial Attention Module (3DSA) for multi-view 3D human pose estimation, which enhances the weights of important features by dividing the voxel feature space into multiple regions and assigning attention weights to each region;
3iGS: Factorised Tensorial Illumination for 3D Gaussian Splatting
Zhe Jun Tang (Nanyang Technological University), Tat-Jen Cham (Nanyang Technological University)
GenerationNeural Radiance FieldGaussian SplattingImage
🎯 What it does: By incorporating continuous illumination fields and Gaussian-specific BRDF features into 3D Gaussian Splatting, a neural renderer is used to achieve more realistic view-dependent specular rendering, significantly improving the original method's performance in reflective and glossy scenarios.
3R-INN: How to be climate friendly while consuming/delivering videos?
ZOUBIDA AMEUR, Daniel Menard
CompressionOptimizationConvolutional Neural NetworkFlow-based ModelImage
🎯 What it does: Achieve a triple processing of video resolution reduction, grain removal, and energy consumption reduction through a single reversible network 3R-INN, reducing video chain energy consumption while enabling recovery of original content.
3x2: 3D Object Part Segmentation by 2D Semantic Correspondences
Anh Thai (Georgia Institute of Technology), Matt Feiszli (Meta AI, FAIR)
SegmentationDiffusion modelImagePoint Cloud
🎯 What it does: Propose a training-agnostic, language-free 3D object part segmentation method 3-By-2, achieving zero-shot and few-shot 3D part segmentation using 2D semantic correspondences and SAM masks.
4D Contrastive Superflows are Dense 3D Representation Learners
Xiang Xu (Nanjing University of Aeronautics and Astronautics), Qingshan Liu (Shanghai AI Laboratory)
Autonomous DrivingRepresentation LearningConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningMultimodalityPoint Cloud
🎯 What it does: Propose the SuperFlow framework, which utilizes continuous LiDAR-camera pairs for spatiotemporal contrastive learning to achieve dense 3D representation learning.
4Diff: 3D-Aware Diffusion Model for Third-to-First Viewpoint Translation
Feng Cheng (FAIR, Meta AI), Kristen Grauman (FAIR, Meta AI)
Image TranslationDepth EstimationTransformerDiffusion modelImagePoint Cloud
🎯 What it does: Propose the 4Diff model to generate first-person perspective images from third-person perspective images;
6DGS: 6D Pose Estimation from a Single Image and a 3D Gaussian Splatting Model
Matteo Bortolon (Fondazione Istituto Italiano di Tecnologia), Alessio Del Bue (Durham University)
Pose EstimationGaussian SplattingImage
🎯 What it does: Propose a single-image 6DoF camera pose estimation method based on the 3D Gaussian Splatting model (6DGS).
6DoF Head Pose Estimation through Explicit Bidirectional Interaction with Face Geometry
Sungho Chun (Kwangwoon University), Ju Yong Chang (Kwangwoon University)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a network called TRG, which simultaneously estimates 6DoF head pose (translation, rotation) and dense 3D facial landmarks from a single image using an explicit bidirectional interaction structure;
A Cephalometric Landmark Regression Method based on Dual-encoder for High-resolution X-ray Image
Chao Dai (Tianjin University), Minpeng Xu (Tianjin University)
Pose EstimationConvolutional Neural NetworkTransformerImageBiomedical Data
🎯 What it does: Propose a dual-encoder regression framework (D-CeLR) that detects all craniofacial measurement landmarks on high-resolution X-ray images in a single pass.
A Closer Look at GAN Priors: Exploiting Intermediate Features for Enhanced Model Inversion Attacks
Yixiang Qiu (Harbin Institute of Technology), Shu-Tao Xia (Tsinghua University)
Adversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: Propose a new GAN-based model inversion attack method called IF-GMI, which decomposes the StyleGAN2 generator and recovers private images through layer-by-layer optimization of intermediate features.
A Compact Dynamic 3D Gaussian Representation for Real-Time Dynamic View Synthesis
Kai Katsumata (University of Tokyo), Hideki Nakayama (University of Tokyo)
GenerationComputational EfficiencyNeural Radiance FieldGaussian SplattingOptical FlowVideo
🎯 What it does: Propose a compact dynamic 3D Gaussian representation, modeling position and rotation with time functions to achieve real-time dynamic view synthesis from monocular/few-view inputs.
A Comparative Study of Image Restoration Networks for General Backbone Network Design
Xiangyu Chen (University of Macau), Chao Dong (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences)
RestorationSuper ResolutionConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: This paper conducts a benchmark comparison of five representative image restoration networks across five classic tasks (super-resolution, denoising, deblurring, deraining, and dehazing), and designs a more task-agnostic backbone network called X-Restormer based on the comparison results.
A Comprehensive Study of Multimodal Large Language Models for Image Quality Assessment
Tianhe Wu (Tsinghua University), Lei Zhang (OPPO Research Institute)
Large Language ModelPrompt EngineeringVision Language ModelImageMultimodalityChain-of-Thought
🎯 What it does: This paper systematically evaluates the performance of multimodal large language models (MLLM) in image quality assessment (IQA), explores nine prompting strategies, and designs a method for selecting difficult samples;
A Diffusion Model for Simulation Ready Coronary Anatomy with Morpho-skeletal Control
Karim Kadry (MIT), Elazer R Edelman
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelAuto EncoderBiomedical DataComputed Tomography
🎯 What it does: This paper proposes a coronary artery anatomy synthesis framework based on latent diffusion models, achieving controllable generation and editing through morphology and skeleton conditions.
A Direct Approach to Viewing Graph Solvability
Federica Arrigoni (Politecnico di Milano), Tomas Pajdla (Czech Technical University in Prague)
Pose EstimationOptimizationGraph
🎯 What it does: This paper proposes a multi-view solvability determination method based on the direct association between the camera matrix and the fundamental matrix, and introduces the concept of 'differential solvability' to efficiently judge solvability and decompose unsolvable graphs.
A Fair Ranking and New Model for Panoptic Scene Graph Generation
Julian Lorenz (University of Augsburg), Rainer Lienhart (University of Augsburg)
Object DetectionSegmentationTransformerPrompt EngineeringImage
🎯 What it does: This paper corrects the evaluation protocol for panoptic scene graph generation (PSGG), proposes the Single Mask Per Object Evaluation Protocol, and re-evaluates existing methods based on this protocol; subsequently, it introduces a decoupled two-stage Transformer architecture called DSFormer, which leverages prior semantic segmentation masks for relation classification and achieves state-of-the-art performance across multiple metrics on the PSOG task.
A Geometric Distortion Immunized Deep Watermarking Framework with Robustness Generalizability
Linfeng Ma (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
Convolutional Neural NetworkTransformerImage
🎯 What it does: Propose a deep watermarking framework robust against both geometric and non-geometric distortions
A Graph-Based Approach for Category-Agnostic Pose Estimation
Or Hirschorn (Tel Aviv University), Shai Avidan (Tel Aviv University)
Pose EstimationGraph Neural NetworkTransformerImageBenchmark
🎯 What it does: Proposed a category-agnostic pose estimation method based on graph neural networks, GraphCape, which treats key points as a graph structure and integrates GCN into the Transformer decoder to transmit structural information; simultaneously updates the skeleton annotations of the MP-100 dataset.
A high-quality robust diffusion framework for corrupted dataset
Quan Dao (VinAI Research), Anh Tran (VinAI Research)
GenerationAnomaly DetectionDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Proposes the first diffusion model framework, RDUOT, capable of maintaining robustness on contaminated datasets.
A Multimodal Benchmark Dataset and Model for Crop Disease Diagnosis
Xiang Liu (China Unicom), Shiguo Lian (China Unicom)
ClassificationRecognitionLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmarkAgriculture Related
🎯 What it does: This paper constructs a multimodal dataset named CDDM, containing 137,000 crop disease images and 1 million question-answer pairs, and employs LoRA for joint fine-tuning of the visual encoder, adapter, and language model to enhance crop disease diagnosis performance.
A New Dataset and Framework for Real-World Blurred Images Super-Resolution
Rui Qin (Tsinghua University), Bin Wang (Kuaishou Technology)
RestorationSuper ResolutionConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Proposes a new ReBlurSR dataset and PBaSR framework to enhance the blind super-resolution quality of images with intentional blur without increasing inference costs.
A Probability-guided Sampler for Neural Implicit Surface Rendering
Gonçalo José Dias Pais (Mitsubishi Electric Research Laboratories), Pedro Miraldo (Toyota Research Institute)
Neural Radiance FieldImage
🎯 What it does: Designed a probability-guided sampler that utilizes the probability density function derived from SDF during the training phase of neural implicit surface rendering. It first samples in the 3D image space (image coordinates + depth), then projects the sampled points into rays, and provides additional samples near each ray. A new surface reconstruction loss is also proposed, combining constraints from near-surface regions and free space.
A Riemannian Approach for Spatiotemporal Analysis and Generation of 4D Tree-shaped Structures
Tahmina Khanam (Murdoch University), Hamid Laga (Murdoch University)
GenerationData SynthesisComputational EfficiencyPoint CloudMeshTime SeriesSequentialAgriculture Related
🎯 What it does: Propose a Riemannian method based on SRVFT for spatiotemporal registration, geometric analysis, statistical modeling, and generation of tree-like 4D structures.
A Rotation-invariant Texture ViT for Fine-Grained Recognition of Esophageal Cancer Endoscopic Ultrasound Images
Tianyi Liu (Southeast University), Yang Chen (Southeast University)
ClassificationRecognitionConvolutional Neural NetworkTransformerContrastive LearningBiomedical DataMagnetic Resonance ImagingUltrasound
🎯 What it does: Propose a rotation-invariant texture visual Transformer (SRRM-ViT) for fine-grained classification of esophageal cancer EUS images, achieving automatic selection of important texture tokens to avoid manual segmentation.
A Secure Image Watermarking Framework with Statistical Guarantees via Adversarial Attacks on Secret Key Networks
Feiyu CHEN, Antoni Chan (City University of Hong Kong)
Safty and PrivacyAdversarial AttackImage
🎯 What it does: Propose a zero-bit watermarking framework based on secret key networks and adversarial attacks, training a network to output a standard multivariate normal distribution. During watermark embedding, the PGD attack is utilized to make the network output a specified vector, achieving detectable and non-replicable watermarks.
A Semantic Space is Worth 256 Language Descriptions: Make Stronger Segmentation Models with Descriptive Properties
Junfei Xiao (Johns Hopkins University), Cihang Xie (University of California, Santa Cruz)
SegmentationExplainability and InterpretabilityTransformerLarge Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Use large language models (LLM) to generate descriptive attributes, constructing an attribute-level label space. Utilize sentence embeddings and K-Means clustering to obtain an interpretable set of attributes. Then, train a semantic segmentation model with multi-label attribute supervision, ultimately outputting attribute activation maps at the pixel level and recovering traditional category predictions through attribute similarity.
A Simple Background Augmentation Method for Object Detection with Diffusion Model
Yuhang Li (Sony AI), Lingjuan Lyu (Sony AI)
Object DetectionPrompt EngineeringDiffusion modelImage
🎯 What it does: Utilize text-to-image diffusion models to perform mask inpainting on the backgrounds of training images, achieving data augmentation and enhancing object detection performance.
A Simple Baseline for Spoken Language to Sign Language Translation with 3D Avatars
Ronglai Zuo (Hong Kong University of Science and Technology), Xin Tong (Microsoft Research Asia)
GenerationPose EstimationTransformerLarge Language ModelVideoTextMesh
🎯 What it does: Proposes a complete baseline system for Spoken2Sign translation using 3D avatars, incorporating lexicon-video dictionary construction, 3D sign estimation based on SMPLSign-X, text-lexicon translation, sign connector, and 3D rendering modules.
A Simple Latent Diffusion Approach for Panoptic Segmentation and Mask Inpainting
Wouter Van Gansbeke (Segments.ai), Bert De Brabandere (Segments.ai)
RestorationSegmentationConvolutional Neural NetworkDiffusion modelAuto EncoderImage
🎯 What it does: Proposing a fully generative panoptic segmentation and mask filling framework based on latent diffusion models
A Simple Low-bit Quantization Framework for Video Snapshot Compressive Imaging
Miao Cao (Zhejiang University), Xin Yuan (Westlake University)
CompressionTransformerVideo
🎯 What it does: Proposes the Q-SCI low-bit quantization framework for end-to-end video snapshot compressive imaging reconstruction, significantly reducing computational cost while maintaining high reconstruction quality.
A Task is Worth One Word: Learning with Task Prompts for High-Quality Versatile Image Inpainting
Junhao Zhuang (Tsinghua Shenzhen International Graduate School Tsinghua University), Kai Chen (Shanghai Artificial Intelligence Laboratory)
RestorationGenerationPrompt EngineeringDiffusion modelImageText
🎯 What it does: Propose PowerPaint, a multi-task image restoration model based on Stable Diffusion, capable of simultaneously performing text-driven object filling, context-aware filling, object removal, and shape-guided filling.
A Unified Anomaly Synthesis Strategy with Gradient Ascent for Industrial Anomaly Detection and Localization
Qiyu Chen (Chinese Academy of Sciences), Zhengtao Zhang (Chinese Academy of Sciences)
Data SynthesisAnomaly DetectionConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Proposed GLASS, a unified strategy for global and local anomaly synthesis, generating anomalies at both feature and image levels through gradient ascent and noise guidance to achieve industrial anomaly detection and localization.
A Unified Image Compression Method for Human Perception and Multiple Vision Tasks
Sha Guo (Peking University), Lingyu Duan (Peking University)
CompressionDiffusion modelAuto EncoderImage
🎯 What it does: A unified image compression framework is studied, which can simultaneously meet human visual perceptual quality and multi-task machine vision performance at extremely low bitrates, and supports open-set (unknown task) scenarios.
A Watermark-Conditioned Diffusion Model for IP Protection
Rui Min (Hong Kong University of Science and Technology), Minhao Cheng (Pennsylvania State University)
GenerationSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: Propose a watermark-conditional generation framework WaDiff based on diffusion models, embedding user-specific watermarks into images during the generation process to achieve IP protection, detection, and user tracking.
ABC Easy as 123: A Blind Counter for Exemplar-Free Multi-Class Class-agnostic Counting
Michael A Hobley, Victor Adrian Prisacariu
Data SynthesisTransformerImage
🎯 What it does: Proposed an example-free multi-class unsupervised counting method named ABC123, and created the first-ever multi-class counting dataset MCAC specifically for this task.
AccDiffusion: An Accurate Method for Higher-Resolution Image Generation
Zhihang Lin (Xiamen University), Rongrong Ji (Tencent)
GenerationDiffusion modelImageText
🎯 What it does: Propose AccDiffusion, a training-free method for generating high-resolution images and effectively eliminating patch-level repetition issues.
Accelerating Image Generation with Sub-path Linear Approximation Model
Chen Xu (Nanjing University), Limin Wang (Nanjing University)
GenerationKnowledge DistillationDiffusion modelScore-based ModelImageOrdinary Differential Equation
🎯 What it does: Propose a Sub-Path Linear Approximation Model (SPLAM), which approximates probabilistic flow (PF-ODE) trajectories using sub-path linear ODEs, and design an efficient SPLAD distillation method to enable pre-trained diffusion models to generate high-quality images with minimal sampling steps.
Accelerating Image Super-Resolution Networks with Pixel-Level Classification
Jinho Jeong (Yonsei University Samsung Advanced Institute of Technology), Seon Joo Kim (Yonsei University Samsung Advanced Institute of Technology)
Super ResolutionImage
🎯 What it does: Proposed the PCSR model, which dynamically allocates computational resources in single-image super-resolution by leveraging pixel-level classification.
Accelerating Online Mapping and Behavior Prediction via Direct BEV Feature Attention
Xunjiang Gu, Boris Ivanovic
SegmentationAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: This paper studies methods for image segmentation using deep learning
Action2Sound: Ambient-Aware Generation of Action Sounds from Egocentric Videos
Changan Chen (University of Texas at Austin), Kristen Grauman (FAIR, Meta)
GenerationDiffusion modelAuto EncoderContrastive LearningVideoMultimodalityRetrieval-Augmented GenerationAudio
🎯 What it does: Proposed a retrieval-enhanced latent diffusion model, AV-LDM, which can generate realistic action audio from silent first-person videos and separate foreground action sounds from background environmental sounds.
ActionSwitch: Class-agnostic Detection of Simultaneous Actions in Streaming Videos
Hyolim Kang (Yonsei University), Seon Joo Kim (Yonsei University)
Object DetectionRecurrent Neural NetworkVideo
🎯 What it does: Proposed the ActionSwitch framework, achieving the first online temporal action localization (On-TAL) method that operates without class information and can detect overlapping actions.
ActionVOS: Actions as Prompts for Video Object Segmentation
Liangyang Ouyang (University of Tokyo), Yoichi Sato (University of Tokyo)
SegmentationTransformerPrompt EngineeringVision-Language-Action ModelVideoTextMultimodality
🎯 What it does: Proposes ActionVOS, a reference video object segmentation framework based on action descriptions, which utilizes human action narratives as language prompts to build a model with a classification head, and introduces an action-aware annotation module and action-guided focal loss, focusing on segmenting active objects.
Active Coarse-to-Fine Segmentation of Moveable Parts from Real Images
Ruiqi Wang (Simon Fraser University), Hao Zhang (Simon Fraser University)
Object DetectionSegmentationPose EstimationTransformerImage
🎯 What it does: This paper proposes a Transformer-based coarse-to-fine stage active learning framework for instance segmentation of movable objects in real indoor RGB images, aiming to minimize manual annotation costs;
Active Generation for Image Classification
Tao Huang (University of Sydney), Chang Xu (University of Sydney)
ClassificationData SynthesisConvolutional Neural NetworkDiffusion modelGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes an image generation framework called ActGen based on active learning, which uses misclassified validation samples to guide the augmentation of training data.
AD3: Introducing a score for Anomaly Detection Dataset Difficulty assessment using VIADUCT dataset
Jan D Lehr (Fraunhofer IPK), Jörg Krüger (TU Berlin)
Anomaly DetectionImageBenchmark
🎯 What it does: Proposed a new industrial anomaly detection dataset called VIADUCT and designed the AD3 score metric to measure dataset difficulty.
AdaCLIP: Adapting CLIP with Hybrid Learnable Prompts for Zero-Shot Anomaly Detection
Yunkang Cao (Huazhong University of Science and Technology), Giacomo Boracchi (Politecnico di Milano)
Anomaly DetectionTransformerPrompt EngineeringVision Language ModelImageBiomedical Data
🎯 What it does: Propose AdaCLIP, a zero-shot anomaly detection method based on CLIP, which significantly improves detection performance by combining static and dynamic learnable prompts.
AdaDiff: Accelerating Diffusion Models through Step-Wise Adaptive Computation
Shengkun Tang, Dongkuan Xu (North Carolina State University)
GenerationComputational EfficiencyConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: Propose the AdaDiff framework, based on the progressive sampling process of diffusion models, dynamically deciding how many network layers to use at each sampling step. Utilize an uncertainty estimation module and uncertainty-weighted hierarchical loss to achieve adaptive early stopping, significantly improving generation speed with almost no loss in image quality.
AdaDiffSR: Adaptive Region-aware Dynamic acceleration Diffusion Model for Real-World Image Super-Resolution
Yuanting Fan (Xi'an Jiaotong University), Xueming Qian (Xi'an Jiaotong University)
Super ResolutionDiffusion modelImage
🎯 What it does: Propose AdaDiffSR, an adaptive dynamic acceleration diffusion model for real-world image super-resolution, which utilizes a multi-metric latent entropy module to dynamically perceive information gain in the latent space, thereby adaptively adjusting denoising steps, and balances the fidelity of the original image with the texture details of the generative model through a progressive feature injection module.