ECCV 2024 Papers — Page 4
European Conference on Computer Vision · 2387 papers
Certifiably Robust Image Watermark
Zhengyuan Jiang (Duke University), Neil Zhenqiang Gong (Pennsylvania State University)
Adversarial AttackImage
🎯 What it does: This paper proposes an image watermarking framework based on randomized smoothing, which provides provably robust watermarking against removal and forgery attacks under L2 norm constraints, and presents three smoothing schemes (multi-class, multi-label, regression) along with derivations and estimations of their robustness lower/upper bounds.
CG-SLAM: Efficient Dense RGB-D SLAM in a Consistent Uncertainty-aware 3D Gaussian Field
Jiarui Hu (Zhejiang University), Zhaopeng Cui (Zhejiang University)
Depth EstimationGaussian SplattingSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: A real-time RGB-D SLAM system based on 3D Gaussian fields (CG-SLAM), achieving dense localization, mapping, and rendering;
Chains of Diffusion Models
Yanheng Wei (Alibaba Group), Shuailei Ma (Northeastern University)
GenerationTransformerVision Language ModelDiffusion modelImageText
🎯 What it does: Propose a multi-condition chain diffusion model framework named Chains, which can gradually generate intermediate conditions such as the number of people, layout, skeleton, 3D structure, and semantic embeddings from text prompts, and finally generate multi-human scene images through ControlNet.
Challenging Forgets: Unveiling the Worst-Case Forget Sets in Machine Unlearning
Chongyu Fan (Michigan State University), Sijia Liu (Michigan State University)
OptimizationSafty and PrivacyImage
🎯 What it does: Proposes identifying the most challenging forgetting set (worst-case forget set) from an adversarial perspective to evaluate the robustness of machine unlearning methods.
Chameleon: A Data-Efficient Generalist for Dense Visual Prediction in the Wild
Donggyun Kim (School of Computing, KAIST), Seunghoon Hong (School of Computing, KAIST)
SegmentationPose EstimationDepth EstimationRepresentation LearningMeta LearningTransformerSupervised Fine-TuningContrastive LearningImageMultimodality
🎯 What it does: Designed and trained a universal model called Chameleon that can adapt to any dense visual prediction task under extremely few samples;
Champ: Controllable and Consistent Human Image Animation with 3D Parametric Guidance
Shenhao Zhu (Nanjing University), Siyu Zhu (Fudan University)
GenerationTransformerVision Language ModelVision-Language-Action ModelDiffusion modelAuto EncoderVideo
🎯 What it does: Using the 3D human parameter model SMPL within a latent diffusion framework, this work provides shape alignment and motion guidance for human image animation, generating spatiotemporally consistent and morphologically realistic videos.
Characterizing Model Robustness via Natural Input Gradients
Adrian Rodriguez-Munoz, Antonio Torralba (MIT)
Explainability and InterpretabilityAdversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper explores and achieves efficient adversarial robustness by regularizing the gradient norm of model inputs on natural samples;
Chat-Edit-3D: Interactive 3D Scene Editing via Text Prompts
shuangkang fang, Ming-Hsuan Yang (Google)
GenerationData SynthesisTransformerLarge Language ModelAgentic AIPrompt EngineeringMixture of ExpertsNeural Radiance FieldGaussian SplattingImageText
🎯 What it does: Propose CE3D, a dialogue-based 3D scene editing framework based on large language models (LLMs), which maps 3D views to a 2D atlas using Hash-Atlas, transforming 3D editing into 2D editing, and automatically invokes multiple visual experts through LLM to achieve arbitrary text interaction and multi-round dialogue.
ChEX: Interactive Localization and Region Description in Chest X-rays
Philip Müller (Technical University of Munich), Daniel Rueckert (Technical University of Munich)
Object DetectionGenerationExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodalityBiomedical Data
🎯 What it does: Propose a multi-task model named ChEX that can generate and locate textual descriptions of lesions or structures in chest X-rays based on text prompts or bounding box queries.
Chronologically Accurate Retrieval for Temporal Grounding of Motion-Language Models
Kent Fujiwara (LY Corporation), Qing Yu (LY Corporation)
RetrievalTransformerLarge Language ModelVision-Language-Action ModelContrastive LearningVideoTextMultimodality
🎯 What it does: Proposes Chronologically Accurate Retrieval (CAR) testing and employs contrastive learning with negative samples generated by shuffling event sequences to enhance the temporal understanding of action-language models, validating the improvements on text-action retrieval and generation tasks.
CIC-BART-SSA: : Controllable Image Captioning with Structured Semantic Augmentation
Kalliopi Basioti (Rutgers University), Afsaneh Fazly (Samsung AI Centre - Toronto)
GenerationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: Developed a structured semantic enhancement (SSA) method based on AMR, automatically generating diverse focused image captions and using them to train the controlled image captioning model CIC-BART-SSA.
CipherDM: Secure Three-Party Inference for Diffusion Model Sampling
Xin Zhao (University of Chinese Academy of Sciences), Zhendong Zhao (Chinese Academy of Sciences)
GenerationSafty and PrivacyDiffusion modelImage
🎯 What it does: Designed and implemented CipherDM, a three-party secure multi-party computation (3PC) framework for secure sampling under diffusion models (DDPM, DDIM, Stable Diffusion), ensuring privacy of model parameters and user inputs.
City-on-Web: Real-time Neural Rendering of Large-scale Scenes on the Web
Kaiwen Song (University of Science and Technology of China), Juyong Zhang (University of Science and Technology of China)
GenerationNeural Radiance FieldImage
🎯 What it does: Achieve real-time neural rendering for large-scale scenes on a Web platform using block partitioning, multi-shader volume rendering, and level of detail (LOD) strategies;
CityGaussian: Real-time High-quality Large-Scale Scene Rendering with Gaussians
Yang Liu (Institute of Automation, Chinese Academy of Sciences), Zhaoxiang Zhang (Institute of Automation, Chinese Academy of Sciences)
GenerationCompressionNeural Radiance FieldGaussian SplattingImagePoint Cloud
🎯 What it does: In CityGaussian, the authors first generate a global rough 3D Gaussian prior, then employ a 'divide and conquer' strategy to split large-scale scenes into multiple blocks for parallel refinement training. Subsequently, they perform multi-level detail compression on the merged global Gaussians, achieving real-time large-scale rendering through block-based detail level selection and aggregation.
CityGuessr: City-Level Video Geo-Localization on a Global Scale
Parth Parag Kulkarni (University of Central Florida), Mubarak Shah (University of Central Florida)
ClassificationTransformerVision Language ModelVideoText
🎯 What it does: This paper proposes a global-scale video geolocation task, aiming to hierarchically predict the city, state/province, country, and continent where the video is located.
CLAMP-ViT: Contrastive Data-Free Learning for Adaptive Post-Training Quantization of ViTs
Akshat Ramachandran (Georgia Institute of Technology), Tushar Krishna (Georgia Institute of Technology)
Data SynthesisCompressionTransformerContrastive LearningImage
🎯 What it does: Propose a data-free post-training quantization method called CLAMP-ViT, specifically designed for unified quantization of weights and activations in visual Transformers (ViT), adaptively generating semantically rich synthetic data during the process to achieve high-precision compression.
CLAP: Isolating Content from Style through Contrastive Learning with Augmented Prompts
Yichao Cai (University of Adelaide), Javen Qinfeng Shi (University of Adelaide)
Domain AdaptationRepresentation LearningData-Centric LearningPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: By performing contrastive learning with image and text augmentation on a pre-trained CLIP model, content features are extracted and style information is removed, enhancing the generalization ability of multi-modal representations.
Class-Agnostic Object Counting with Text-to-Image Diffusion Model
Xiaofei Hui (Lancaster University), Jun Liu (Lancaster University)
Object DetectionGenerationConvolutional Neural NetworkTransformerSupervised Fine-TuningPrompt EngineeringDiffusion modelImage
🎯 What it does: This paper proposes the CountDiff framework, which utilizes cross-attention and self-attention mechanisms from the pre-trained text-to-image diffusion model (Stable Diffusion) to achieve few-shot and zero-shot category-agnostic object counting.
Class-Incremental Learning with CLIP: Adaptive Representation Adjustment and Parameter Fusion
Linlan Huang (Nankai University), Xialei Liu (Nankai University)
Representation LearningTransformerSupervised Fine-TuningVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes the RAPF method, achieving class-incremental learning on the CLIP model through text-guided neighborhood class separation and parameter fusion, significantly reducing catastrophic forgetting.
Classification Matters: Improving Video Action Detection with Class-Specific Attention
Jinsung Lee (Pohang University of Science and Technology), Suha Kwak (Pohang University of Science and Technology)
RecognitionTransformerVideo
🎯 What it does: This paper addresses the classification bottleneck in video action detection by proposing a Transformer framework based on class-specific attention;
Clean & Compact: Efficient Data-Free Backdoor Defense with Model Compactness
Huy Phan (Rutgers University), Bo Yuan (Rutgers University)
Knowledge DistillationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Proposes the Clean & Compact (C&C) method, which removes backdoors and achieves model compression without training data by leveraging singular value rank sensitivity.
ClearCLIP: Decomposing CLIP Representations for Dense Vision-Language Inference
Mengcheng Lan (S-Lab Nanyang Technological University), Wayne Zhang (SenseTime Research)
SegmentationTransformerVision Language ModelContrastive LearningImage
🎯 What it does: Investigated the root causes of noise generated by CLIP in open-source semantic segmentation tasks, proposing ClearCLIP which significantly improves segmentation quality by removing residual connections, adopting self-attention mechanisms, and eliminating FFN.
CLEO: Continual Learning of Evolving Ontologies
Shishir Muralidhara (German Research Center for Artificial Intelligence), René Schuster
SegmentationKnowledge DistillationRepresentation LearningImage
🎯 What it does: Proposed CLEO (Continual Learning Evolving Ontology), a novel incremental learning setup aimed at addressing the issue of class evolution in continual learning.
Click Prompt Learning with Optimal Transport for Interactive Segmentation
Jie Liu (University of Amsterdam), Efstratios Gavves (University of Amsterdam)
SegmentationTransformerPrompt EngineeringImageBiomedical Data
🎯 What it does: Developed an interactive image segmentation model called CPlot based on click prompt learning, which utilizes optimal transport optimization to refine click prompts and capture diverse user intentions.
Click-Gaussian: Interactive Segmentation to Any 3D Gaussians
Seokhun Choi (LG Electronics), Hoseok Do (LG Electronics)
SegmentationContrastive LearningGaussian SplattingImage
🎯 What it does: This paper proposes Click-Gaussian, a fast interactive segmentation method based on 3D Gaussian Splatting, which achieves precise segmentation from coarse to fine using two-level granularity feature fields.
CLIFF: Continual Latent Diffusion for Open-Vocabulary Object Detection
Wuyang Li (Chinese University of Hong Kong), Yixuan Yuan (Chinese University of Hong Kong)
Object DetectionVision Language ModelDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: This paper proposes a continuous latent diffusion framework named CLIFF for open-vocabulary object detection (OVD), which enhances the detection performance of novel categories by enabling continuous distribution transfer between object, image, and text latent spaces.
CliffPhys: Camera-based Respiratory Measurement using Clifford Neural Networks
Omar Ghezzi (Università degli Studi di Milano), Alessandro D'Amelio (Università degli Studi di Milano)
Convolutional Neural NetworkSupervised Fine-TuningOptical FlowVideoMultimodality
🎯 What it does: Developed a Camera-based Respiratory Measurement model called CliffPhys based on Clifford neural networks, which utilizes optical flow and monocular depth estimation to obtain 2D vector fields and scalar fields, and captures geometric relationships through Clifford Neural Layers to achieve non-contact breathing rate measurement.
CLIP-DINOiser: Teaching CLIP a few DINO tricks for open-vocabulary semantic segmentation
Monika Wysoczańska (Warsaw University Of Technology), Patrick Pérez (valeo.ai)
SegmentationConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningImage
🎯 What it does: Propose a lightweight CLIP-DINOiser method that leverages self-supervised DINO feature similarity information to guide MaskCLIP's local feature aggregation and learns CLIP's own background mask, thereby improving open-vocabulary semantic segmentation performance without requiring any annotations.
CLIP-DPO: Vision-Language Models as a Source of Preference for Fixing Hallucinations in LVLMs
Yassine Ouali (Samsung AI Center), Georgios Tzimiropoulos (Queen Mary University of London)
OptimizationTransformerSupervised Fine-TuningReinforcement LearningContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Rank descriptions generated by LVLM using a pre-trained CLIP model to construct positive/negative pairs, then perform fine-tuning with Direct Preference Optimization (DPO) to reduce hallucination rates in image-language models.
CLIP-Guided Generative Networks for Transferable Targeted Adversarial Attacks
Hao Fang (Tsinghua University), Shu-Tao Xia (Tsinghua University)
Adversarial AttackTransformerGenerative Adversarial NetworkImageText
🎯 What it does: This paper proposes a generative network called CGNC that leverages CLIP text knowledge through a cross-attention module to achieve multi-target transferable directed adversarial attacks;
Closed-Loop Unsupervised Representation Disentanglement with $\\beta$-VAE Distillation and Diffusion Probabilistic Feedback
Xin Jin (Ningbo Institute of Digital Twin, Eastern Institute of Technology), Wenjun Zeng (Xi'an Jiaotong University)
GenerationKnowledge DistillationRepresentation LearningDiffusion modelAuto EncoderOptical FlowImage
🎯 What it does: Proposed a closed-loop unsupervised representation disentanglement framework called CL-Dis, combining diffusion autoencoders (Diff-AE) with β-VAE, utilizing knowledge distillation and diffusion feedback to achieve mutually reinforcing representation learning, and identifying disentangled dimensions through self-supervised navigation methods.
CLOSER: Towards Better Representation Learning for Few-Shot Class-Incremental Learning
Junghun Oh (Seoul National University), Kyoung Mu Lee (Seoul National University)
ClassificationRepresentation LearningContrastive LearningImage
🎯 What it does: This paper addresses the representation learning problem in few-shot class-incremental learning (FSCIL) by proposing a learning strategy that balances discriminability and transferability during training on base classes. The core idea is to jointly train using three losses: low-temperature cross-entropy, contrastive learning, and cross-class distance minimization, which bring features of different classes closer together, forming a compressed yet well-dispersed representation space.
CloudFixer: Test-Time Adaptation for 3D Point Clouds via Diffusion-Guided Geometric Transformation
Hajin Shim (Korea Advanced Institute of Science and Technology), Eunho Yang (Korea Advanced Institute of Science and Technology)
Domain AdaptationDiffusion modelPoint Cloud
🎯 What it does: Implement test-time adaptation (TTA) for 3D point cloud inference in distribution drift scenarios by enabling input adaptation during testing through geometric transformations guided by a pre-trained diffusion model, significantly enhancing model robustness.
CLR-GAN: Improving GANs Stability and Quality via Consistent Latent Representation and Reconstruction
Shengke Sun (Tiangong University), Shuzhen Han (Tiangong University)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Propose a new GAN training paradigm called CLR-GAN, treating the generator and discriminator as inverse processes. Introduce a latent space reconstruction task in the discriminator and a real image reconstruction task in the generator, forming latent space consistency loss and reconstruction loss to enhance GAN training stability and generation quality.
ClusteringSDF: Self-Organized Neural Implicit Surfaces for 3D Decomposition
Tianhao Wu (Nanyang Technological University), Tat-Jen Cham (Nanyang Technological University)
SegmentationNeural Radiance FieldImageMeshBenchmark
🎯 What it does: This paper proposes an unsupervised 3D scene segmentation and reconstruction method called ClusteringSDF, which can self-organize multi-view inconsistent 2D semantic/instance segmentation information into a consistent 3D asset segmentation through neural implicit surfaces (SDF), while simultaneously reconstructing scene geometry.
CMD: A Cross Mechanism Domain Adaptation Dataset for 3D Object Detection
Jinhao Deng (Xiamen University), Cheng Wang (Xiamen University)
Object DetectionDomain AdaptationConvolutional Neural NetworkMultimodalityPoint CloudBenchmark
🎯 What it does: This paper proposes a cross-mechanism domain adaptation dataset named CMD, and builds a three-stage domain adaptation baseline method called DIG based on this dataset, aiming to address the performance degradation in point cloud detection across different LiDAR, radar, and camera sensors.
CMTA: Cross-Modal Temporal Alignment for Event-guided Video Deblurring
Taewoo Kim (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
RestorationTransformerVideoMultimodalityBenchmark
🎯 What it does: Proposes a video deblurring method (CMTA) that leverages the high temporal resolution information of event cameras, comprising two major modules: Cross-Modal Recursive Internal Feature Enhancement (CRIFE) and Event-Guided Cascaded Cross-Frame Temporal Feature Alignment (ECITFA), and constructs a novel real-world hybrid camera dataset named EVRB.
Co-speech Gesture Video Generation with 3D Human Meshes
Aniruddha Mahapatra, Jing Xiao (PAII Inc)
GenerationVision-Language-Action ModelAuto EncoderGenerative Adversarial NetworkVideoMeshAudio
🎯 What it does: Designed an audio-driven co-speech gesture video generation pipeline, which first predicts facial, body, and hand motions using 3D human meshes (SMPL-X), and then generates realistic facial expressions and gesture videos through UV texture optimization and conditional GAN.
Co-Student: Collaborating Strong and Weak Students for Sparsely Annotated Object Detection
Lianjun Wu (Huazhong University of Science and Technology), Xinggang Wang (Jingce Electronic Group Co Ltd)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Proposed the Co-Student framework, which utilizes teacher-student dual students to collaboratively generate and denoise pseudo labels, thereby fully leveraging unlabeled objects in sparse annotation object detection (SAOD) tasks.
Co-synthesis of Histopathology Nuclei Image-Label Pairs using a Context-Conditioned Joint Diffusion Model
Seonghui Min (Korea University), Won-Ki Jeong (Korea University)
GenerationData SynthesisPrompt EngineeringVision Language ModelDiffusion modelMultimodalityBiomedical Data
🎯 What it does: A context-conditioned joint diffusion model is used to simultaneously synthesize cell nucleus images, semantic labels, and distance maps in pathological tissue sections, achieving one-click generation of images and labels.
Coarse-to-Fine Implicit Representation Learning for 3D Hand-Object Reconstruction from a Single RGB-D Image
Xingyu Liu (Beijing University of Posts and Telecommunications), Jianxin Liao (Beijing University of Posts and Telecommunications)
Depth EstimationRepresentation LearningConvolutional Neural NetworkNeural Radiance FieldImagePoint Cloud
🎯 What it does: This paper proposes a coarse-to-fine implicit surface learning framework that progressively constructs the SDF of hand-object interactions using RGB-D inputs;
Cocktail Universal Adversarial Attack on Deep Neural Networks
Shaoxin Li (Chongqing University), Lingyang Chu (McMaster University)
ClassificationAdversarial AttackImage
🎯 What it does: Propose a 'cocktail' general adversarial attack framework that utilizes K diverse universal adversarial perturbations (UAP) and selects the most effective perturbation for each new image through a selection network to achieve the attack;
COD: Learning Conditional Invariant Representation for Domain Adaptation Regression
Hao-Ran Yang (Sun Yat-Sen University), You-Wei Luo (Sun Yat-Sen University)
Domain AdaptationRepresentation LearningBenchmark
🎯 What it does: Proposes an unsupervised domain adaptation regression method based on conditional operator discrepancy (COD), establishing a theoretical framework for conditional distribution alignment with continuous labels and providing a measurable COD statistic;
CoDA: Instructive Chain-of-Domain Adaptation with Severity-Aware Visual Prompt Tuning
ZiYang Gong, Zhenming Ji (Sun Yat-sen University)
SegmentationDomain AdaptationAutonomous DrivingPrompt EngineeringVision Language ModelImageChain-of-Thought
🎯 What it does: Propose the CoDA method, combining Chain-of-Domain phased adaptation and Severity-Aware Visual Prompt Tuning to achieve unsupervised domain adaptation for adverse scenarios;
CogView3: Finer and Faster Text-to-Image Generation via Relay Diffusion
Wendi Zheng (Tsinghua University), Jie Tang (Tsinghua University)
GenerationKnowledge DistillationTransformerDiffusion modelAuto EncoderImageTextOrdinary Differential Equation
🎯 What it does: Propose CogView3, a two-stage latent space text-to-image generation model based on relay diffusion, achieving fast synthesis of high-resolution (up to 2048×2048) images.
CoherentGS: Sparse Novel View Synthesis with Coherent 3D Gaussians
Avinash Paliwal (Texas A&M University), Nima Khademi Kalantari (Meta Reality Labs)
GenerationConvolutional Neural NetworkNeural Radiance FieldGaussian SplattingOptical FlowImage
🎯 What it does: Achieving novel view synthesis using 3D Gaussian Splatting under sparse viewpoints.
COHO: Context-Sensitive City-Scale Hierarchical Urban Layout Generation
Liu He (Purdue University), Daniel Aliaga (Purdue University)
GenerationGraph Neural NetworkAuto EncoderGraph
🎯 What it does: Propose a graph-based masked autoencoder (GMAE) and priority scheduling method to generate large-scale city-level 2.5D layouts, considering multi-level context sensitivity.
COIN-Matting: Confounder Intervention for Image Matting
Zhaohe Liao (Shanghai Jiao Tong University), Liqing Zhang (Shanghai Jiao Tong University)
SegmentationImage
🎯 What it does: This paper proposes a model-agnostic image matting framework called COIN based on causal intervention, which improves matting quality by eliminating biases in contrast and transparency.
COIN: Control-Inpainting Diffusion Prior for Human and Camera Motion Estimation
Jiefeng Li (NVIDIA), Umar Iqbal (Shanghai Jiao Tong University)
Pose EstimationDiffusion modelScore-based ModelSimultaneous Localization and MappingVideoPoint Cloud
🎯 What it does: This paper proposes a framework called COIN based on a control-inpainting diffusion prior, which is used to simultaneously recover global human motion and camera motion from dynamic camera videos.
CoLA: Conditional Dropout and Language-driven Robust Dual-modal Salient Object Detection
Shuang Hao (Huazhong University of Science and Technology), He Tang (Huazhong University of Science and Technology)
Object DetectionConvolutional Neural NetworkPrompt EngineeringVision Language ModelMultimodality
🎯 What it does: Designed and implemented the CoLA framework to simultaneously address noise input and modality missing issues in dual-modal salient object detection, incorporating two modules: Language-Driven Quality Assessment (LQA) and Conditional Dropout (CD).
CoLeaF: A Contrastive-Collaborative Learning Framework for Weakly Supervised Audio-Visual Video Parsing
Faegheh Sardari (University of Surrey), Adrian Hilton (University of Surrey)
SegmentationKnowledge DistillationTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: Propose the CoLeaF framework for weakly supervised audio-visual video parsing (AVVP), optimizing cross-modal context fusion in the embedding space through contrastive and collaborative learning, specifically distinguishing alignable and non-alignable events.
Collaborative Control for Geometry-Conditioned PBR Image Generation
Shimon Vainer (Unity Technologies), Simon Donné (Unity Technologies)
GenerationData SynthesisVision Language ModelDiffusion modelAuto EncoderImageTextMesh
🎯 What it does: This paper proposes the Collaborative Control framework, which jointly pre-trains RGB diffusion models and a new PBR diffusion model to achieve PBR image generation under geometric conditions.
Collaborative Vision-Text Representation Optimizing for Open-Vocabulary Segmentation
Siyu Jiao (Peng Cheng Laboratory), Humphrey Shi (Georgia Institute of Technology)
SegmentationRepresentation LearningTransformerMultimodality
🎯 What it does: Studies how to co-optimize CLIP visual and text representations to enhance open-vocabulary segmentation performance, proposing the MAFT+ framework with content-dependent transfer and representation compensation.
ColorMAE: Exploring data-independent masking strategies in Masked AutoEncoders
Carlos Hinojosa (King Abdullah University of Science and Technology), Bernard Ghanem (King Abdullah University of Science and Technology)
ClassificationObject DetectionSegmentationRepresentation LearningTransformerAuto EncoderImage
🎯 What it does: Proposes ColorMAE, a self-supervised pre-training method that utilizes noise filtering to generate data-agnostic masks, and applies it to MAE for image reconstruction.
ColorMNet: A Memory-based Deep Spatial-Temporal Feature Propagation Network for Video Colorization
Yixin Yang (Nanjing University of Science and Technology), Jinshan Pan (Nanjing University of Science and Technology)
GenerationConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: Propose a memory-based deep spatiotemporal feature propagation network, ColorMNet, for video degrayscale.
ColorPeel: Color Prompt Learning with Diffusion Models via Color and Shape Disentanglement
Muhammad Atif Butt (Computer Vision Center), Joost van de Weijer (Computer Vision Center)
GenerationPrompt EngineeringDiffusion modelImageMesh
🎯 What it does: Propose the ColorPeel method, which utilizes generated colored geometric shapes to learn precise color prompts, achieving decoupling of color and shape;
COM Kitchens: An Unedited Overhead-view Procedural Videos Dataset a Vision-Language Benchmark
Atsushi Hashimoto (OMRON SINIC X Corp.), Yoshitaka Ushiku (Cookpad Inc.)
RetrievalTransformerVision Language ModelVideoTextBenchmark
🎯 What it does: This paper collects and constructs the COM Kitchens dataset, which includes 40 hours of 145 uncut high-angle cooking videos, and provides fine-grained visual action maps along with corresponding text instructions.
Combining Generative and Geometry Priors for Wide-Angle Portrait Correction
Lan Yao, Wangmeng Zuo (Harbin Institute Of Technology)
RestorationConvolutional Neural NetworkGenerative Adversarial NetworkOptical FlowImage
🎯 What it does: Propose a dual-modal framework based on generative adversarial networks (GANs) that leverages facial priors and geometric symmetry priors to correct facial distortions and background line warping in wide-angle portraits.
ComboVerse: Compositional 3D Assets Creation Using Spatially-Aware Diffusion Guidance
Yongwei Chen (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationDepth EstimationOptimizationDiffusion modelScore-based ModelImageTextMeshBenchmark
🎯 What it does: This paper proposes ComboVerse, a two-stage 3D asset generation framework. First, the input image is processed through target segmentation and occlusion repair, and then a single-target 3D reconstruction model generates a 3D mesh for each target. Subsequently, a spatially-aware diffusion guidance (Spatially-Aware Score Distillation Sampling, SSDS) jointly optimizes the scale, rotation, and displacement of each target to produce a composite 3D asset that conforms to the original image's spatial relationships.
ComFusion: Enhancing Personalized Generation by Instance-Scene Compositing and Fusion
Yan Hong (Ant Group), Jianfu Zhang (Shanghai Jiao Tong University)
GenerationSupervised Fine-TuningVision Language ModelDiffusion modelImageText
🎯 What it does: Propose the ComFusion method, achieving personalized text-to-image generation with a small number of instances (≤5 images) through instance-scene synthesis and fusion.
Common Sense Reasoning for Deep Fake Detection
Yue Zhang (Michigan State University), Gaurav Bharaj (Reality Defender Inc)
Anomaly DetectionExplainability and InterpretabilityTransformerVision Language ModelContrastive LearningImageVideoMultimodality
🎯 What it does: Convert the deepfake detection task into a visual question answering (DD-VQA) task, construct a corresponding dataset, and train a BLIP-based multimodal Transformer model to generate authenticity judgments and textual explanations.
Commonly Interesting Images
Fitim Abdullahu (Zurich University of Applied Sciences), Helmut Grabner (Zurich University of Applied Sciences)
RetrievalRecommendation SystemTransformerVision Language ModelContrastive LearningImage
🎯 What it does: Built a 'shared interest' image evaluation framework based on Flickr user likes behavior, using this definition to mine common and subjective interest images.
COMO: Compact Mapping and Odometry
Eric Dexheimer (Imperial College London), Andrew Davison
Depth EstimationOptimizationSimultaneous Localization and MappingOptical FlowImagePoint Cloud
🎯 What it does: Propose a real-time monocular mapping and odometry system (COMO), which encodes and decodes dense geometry through a set of compact 3D anchors and depth covariance functions, achieving joint optimization of camera pose and dense geometry.
CoMo: Controllable Motion Generation through Language Guided Pose Code Editing
Yiming Huang (University of Pennsylvania), Lingjie Liu (University of Pennsylvania)
GenerationTransformerLarge Language ModelAuto EncoderVideoTextMultimodality
🎯 What it does: The paper proposes CoMo, a language-guided controllable motion generation model that can generate and edit human motion based on text descriptions.
Compact 3D Scene Representation via Self-Organizing Gaussian Grids
Wieland Morgenstern (Fraunhofer Heinrich Hertz Institute), Peter Eisert (Fraunhofer Heinrich Hertz Institute)
CompressionGaussian Splatting
🎯 What it does: Rearrange the high-dimensional parameters of 3D Gaussian Splatting into a locally smooth 2D grid and introduce smooth regularization during training to achieve high-quality compressed scene representations.
Compensation Sampling for Improved Convergence in Diffusion Models
Hui Lu (Utrecht University), Ronald Poppe (Utrecht University)
RestorationGenerationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes a Compensation Sampling technique, which uses a lightweight U-Net to learn compensation terms that guide the inverse process of diffusion models, significantly accelerating training and inference while improving image quality in unconditional generation, facial restoration, and occlusion removal tasks.
CompGS: Smaller and Faster Gaussian Splatting with Vector Quantization
K L Navaneet (University of California, Davis), Hamed Pirsiavash (University of California, Davis)
CompressionComputational EfficiencyGaussian Splatting
🎯 What it does: This paper proposes CompGS, a method that significantly compresses model storage and inference costs while maintaining the real-time rendering and high quality of 3D Gaussian Splatting (3DGS).
COMPOSE: Comprehensive Portrait Shadow Editing
Andrew Z Hou, Xiaoming Liu (Adobe Research)
Image HarmonizationTransformerDiffusion modelAuto EncoderImage
🎯 What it does: Proposed a single-image portrait shadow editing framework named COMPOSE, capable of precisely controlling shadow position, intensity, and shape while preserving the original environmental illumination;
Compositional Substitutivity of Visual Reasoning for Visual Question Answering
Chuanhao Li (Beijing Institute of Technology), Yunde Jia (Beijing Institute of Technology)
TransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: This paper investigates compositional substitutability in visual question answering and proposes a training framework based on a support set, enabling the model to recognize and maintain invariant representations for synonymous primitives (words, visual entities, references).
Comprehensive Attribution: Inherently Explainable Vision Model with Feature Detector
Xianren Zhang (Pennsylvania State University), Suhang Wang (Pennsylvania State University)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: Proposed a novel inherently interpretable visual model called COMET, which utilizes a selector to generate attribution maps and collaborates with a predictor, while incorporating a pre-trained feature detector to enhance attribution completeness and prevent interlocking between the selector and predictor.
Compress3D: a Compressed Latent Space for 3D Generation from a Single Image
Bowen Zhang (Xi'an Jiaotong University), Xi Zhao (Xi'an Jiaotong University)
GenerationConvolutional Neural NetworkDiffusion modelAuto EncoderImagePoint CloudMesh
🎯 What it does: Propose a single-image 3D generation framework based on triplane autoencoders and diffusion models, capable of compressing 3D models into a low-dimensional latent space and performing conditional generation using shape and image embeddings in this space.
Computing the Lipschitz constant needed for fast scene recovery from CASSI measurements
Niels Chr Overgaard (Lund University), Anders Holst (Lund University)
RestorationComputational EfficiencyImage
🎯 What it does: Proposed an efficient method to compute the gradient Lipschitz constant under CASSI measurements, avoiding computational bottlenecks caused by directly calculating eigenvalues of large matrices
CoMusion: Towards Consistent Stochastic Human Motion Prediction via Motion Diffusion
Jiarui Sun (University of Illinois Urbana Champaign), Girish Chowdhary (University of Illinois Urbana Champaign)
GenerationGraph Neural NetworkTransformerDiffusion modelTime SeriesSequential
🎯 What it does: Proposes CoMusion, a single-stage end-to-end diffusion model for consistency-based random human motion prediction.
Concept Arithmetics for Circumventing Concept Inhibition in Diffusion Models
Vitali Petsiuk (Boston University), Kate Saenko (Boston University)
GenerationAdversarial AttackPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper studies the use of concept arithmetic and multi-prompt reasoning in text-to-image diffusion models to bypass existing concept suppression methods and restore suppressed target concepts.
Concept Sliders: LoRA Adaptors for Precise Control in Diffusion Models
Rohit Gandikota (Northeastern University), David Bau (Northeastern University)
GenerationVision Language ModelDiffusion modelImageTextMultimodalityStochastic Differential Equation
🎯 What it does: Designed a LoRA-based concept slider (Concept Sliders) that enables continuous, interpretable fine-grained control over text or visual concepts in diffusion models, and supports multi-attribute editing by synthesizing multiple sliders.
ConceptExpress: Harnessing Diffusion Models for Single-image Unsupervised Concept Extraction
Shaozhe Hao (University of Hong Kong), Kwan-Yee K. Wong (University of Hong Kong)
RecognitionTransformerDiffusion modelImage
🎯 What it does: This paper proposes a method that automatically extracts and reproduces individual concepts from a single multi-concept image under unlabeled conditions using a pre-trained diffusion model.
Conceptual Codebook Learning for Vision-Language Models
Yi Zhang (Harbin Institute of Technology), Zhihai He (University of Missouri)
Representation LearningLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelContrastive LearningImageText
🎯 What it does: Fine-tune CLIP as CoCoLe by introducing a learnable concept codebook and a manually crafted concept cache to enhance cross-domain generalization under few-shot scenarios.
Concise Plane Arrangements for Low-Poly Surface and Volume Modelling
Raphael Sulzer (Centre Inria d'Université Côte d'Azur), Florent Lafarge (Centre Inria d'Université Côte d'Azur)
Computational EfficiencyPoint CloudMesh
🎯 What it does: This paper proposes a scalable planar arrangement method for constructing low-polygon surface and volume models from point clouds.
CONDA: Condensed Deep Association Learning for Co-Salient Object Detection.
Long Li (Northwestern Polytechnical University), Fahad Shahbaz Khan (Mohamed bin Zayed University of Artificial Intelligence)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Proposed the CONDA framework, which leverages deep networks to transform raw pixel-level associations into deep association features, achieving high-quality co-occurrent object detection through evolutionary association generation, association condensation, and object-aware cycle-consistent loss.
ConDense: Consistent 2D-3D Pre-training for Dense and Sparse Features from Multi-View Images
Xiaoshuai Zhang (University Of California San Diego), Leonidas Guibas (Stanford University)
RetrievalRepresentation LearningConvolutional Neural NetworkTransformerNeural Radiance FieldContrastive LearningImageMultimodality
🎯 What it does: Propose the ConDense framework, which jointly performs 2D-3D pre-training using existing 2D pre-trained networks and large-scale multi-view image data, obtaining densely and sparsely co-embedded 2D and 3D features, and achieving 2D-3D consistency constraints through NeRF-style ray marching.
Confidence Self-Calibration for Multi-Label Class-Incremental Learning
Kaile Du (Southeast University), Guangcan Liu (Southeast University)
ClassificationKnowledge DistillationRepresentation LearningGraph Neural NetworkImage
🎯 What it does: To address the challenge of partial labels in multi-label class incremental learning, the Confidence Self-Calibration (CSC) framework is proposed, which mitigates false positives and catastrophic forgetting caused by overconfidence through calibrating label relationships and multi-label confidence.
Confidence-Based Iterative Generation for Real-World Image Super-Resolution
Jialun Peng (University of Science and Technology of China), Dong Liu (University of Science and Technology of China)
GenerationSuper ResolutionTransformerAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: Propose RealSRT, a Transformer-based iterative generation model that can adaptively generate high-resolution images through multi-step iterations.
ConGeo: Robust Cross-view Geo-localization across Ground View Variations
Li Mi (EPFL), Devis Tuia (EPFL)
RetrievalContrastive LearningImage
🎯 What it does: Studied the cross-view geolocation problem and proposed ConGeo, a robust contrastive learning framework that can adapt to variations in ground perspectives (direction and field of view) with a single model.
Connecting Consistency Distillation to Score Distillation for Text-to-3D Generation
Zongrui Li (Nanyang Technological University), Xudong Jiang (Nanyang Technological University)
GenerationData SynthesisDiffusion modelScore-based ModelGaussian SplattingTextMeshOrdinary Differential Equation
🎯 What it does: Building upon existing 3D generation frameworks based on Score Distillation Sampling (SDS), the authors propose Guided Consistency Sampling (GCS) and Brightness-Equalized Generation (BEG), which significantly enhance the detail and realism of text-driven 3D assets by integrating the concept of consistency distillation.
Consistent 3D Line Mapping
Xulong Bai (Chinese Academy of Sciences), Shuhan Shen (Chinese Academy of Sciences)
OptimizationImage
🎯 What it does: Proposes a complete system for reconstructing 3D line segments and their line trajectories from multi-view known camera poses, emphasizing consistency and completeness.
Constructing Concept-based Models to Mitigate Spurious Correlations with Minimal Human Effort
Jeeyung Kim (Purdue University), Qiang Qiu (Purdue University)
Explainability and InterpretabilityData-Centric LearningLarge Language ModelVision Language ModelContrastive LearningImageMultimodalityBenchmark
🎯 What it does: Established a method to construct a concept bottleneck model (CBM) using multimodal large language models and visual tools, requiring almost no manual annotation, thereby reducing spurious correlations in datasets.
Content-Aware Radiance Fields: Aligning Model Complexity with Scene Intricacy Through Learned Bitwidth Quantization
Weihang Liu (ShanghaiTech University), Xin Lou (ShanghaiTech University)
Computational EfficiencyNeural Radiance FieldImage
🎯 What it does: This paper proposes a content-aware radiance fields framework that automatically adjusts the model complexity for each scene by learning learned bitwidth quantization (LBQ) and adversarial content-aware quantization (A-CAQ), enabling high bitwidth for complex scenes and low bitwidth for simple scenes, thereby significantly reducing computational cost while maintaining rendering quality.
Context Diffusion: In-Context Aware Image Generation
Ivona Najdenkoska (Meta GenAI), Filip Radenovic (Meta GenAI)
GenerationTransformerVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Proposes Context Diffusion, a diffusion framework that can simultaneously utilize visual context images and text prompts in image generation, and supports few-shot learning with multiple context images.
Context-Aware Action Recognition: Introducing a Comprehensive Dataset for Behavior Contrast
Tatsuya Sasaki (Hitachi Ltd R&D Group), Satoshi Kondo (Muroran Institute of Technology)
RecognitionData SynthesisConvolutional Neural NetworkRecurrent Neural NetworkLarge Language ModelVision-Language-Action ModelVideoBenchmark
🎯 What it does: This study constructs a video dataset containing comparisons of correct and incorrect industrial behaviors and sports behaviors, recording high-resolution videos from multiple perspectives (fixed cameras, drones, GoPro), with video-level and frame-level annotations for each clip.
Context-Guided Spatial Feature Reconstruction for Efficient Semantic Segmentation
Zhenliang Ni (Huawei Noah's Ark Lab), Yunhe Wang (Huawei Noah's Ark Lab)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: Proposes a lightweight semantic segmentation framework named CGRSeg based on context-guided spatial feature reconstruction.
Contextual Correspondence Matters: Bidirectional Graph Matching for Video Summarization
Yunzuo Zhang (Shijiazhuang Tiedao University), Yameng Liu (Shijiazhuang Tiedao University)
GenerationConvolutional Neural NetworkGraph Neural NetworkTransformerVideo
🎯 What it does: Propose an end-to-end video summarization framework named Bgm4Video based on bidirectional graph matching, which can simultaneously model coarse-grained and fine-grained temporal contexts. By leveraging a graph matching mechanism, the framework enables mutual information transfer among multi-granularity features, thereby generating more accurate and informative video summaries.
Continual Learning and Unknown Object Discovery in 3D Scenes via Self-Distillation
Mohamed El Amine Boudjoghra (Mohamed bin Zayed University of Artificial Intelligence), Fahad Shahbaz Khan (Mohamed bin Zayed University of Artificial Intelligence)
Anomaly DetectionKnowledge DistillationTransformerPoint CloudMesh
🎯 What it does: Proposes the OpenDistill3D method, which utilizes models from previous tasks for self-distillation to jointly address continuous learning and unknown object discovery in 3D scenes, while avoiding the use of sample memory.
Continual Learning for Remote Physiological Measurement: Minimize Forgetting and Simplify Inference
Qian Liang (University of Science and Technology of China), Yang Hu (University of Science and Technology of China)
Domain AdaptationMeta LearningTransformerSupervised Fine-TuningVideoBiomedical DataBenchmark
🎯 What it does: This paper proposes a continual learning framework ADPP for remote optical pulse (rPPG) measurement, addressing the issues of catastrophic forgetting and domain shift during multi-domain sequence learning.
Continuity Preserving Online CenterLine Graph Learning
Yunhui Han (Xiaomi EV), Zhiwei Li (Xiaomi EV)
Autonomous DrivingRecurrent Neural NetworkGraph Neural NetworkPoint Cloud
🎯 What it does: This paper proposes CGNet, an end-to-end continuity-preserving centerline graph learning network for online construction of coherent centerline graphs applicable to advanced driving.
Continuous Memory Representation for Anomaly Detection
Joo Chan Lee (Sungkyunkwan University), Jong Hwan Ko (Sungkyunkwan University)
Anomaly DetectionAuto EncoderImage
🎯 What it does: This paper proposes an unsupervised anomaly detection method called CRAD based on continuous memory, which maps image features to a continuous grid to reconstruct normal features and detects anomalies through difference detection.
Continuous SO(3) Equivariant Convolution for 3D Point Cloud Analysis
Jaein Kim (Seoul National University), Byoung-Tak Zhang (Seoul National University)
ClassificationRetrievalComputational EfficiencyConvolutional Neural NetworkPoint Cloud
🎯 What it does: Developed a continuous SO(3) equivariant convolution layer called CSEConv for 3D point cloud analysis, ensuring rotational equivariance and achieving efficient computation.
Contourlet Residual for Prompt Learning Enhanced Infrared Image Super-Resolution
Xingyuan Li, Risheng Liu (Dalian University of Technology)
Super ResolutionConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelImageText
🎯 What it does: Propose an infrared image super-resolution network combining Contourlet residual and prompt learning, which extracts high-frequency details through multi-scale and multi-directional Contourlet decomposition and achieves semantic-level optimization by guiding the model with positive and negative text prompts.
Contrasting Deepfakes Diffusion via Contrastive Learning and Global-Local Similarities
Lorenzo Baraldi (University of Modena and Reggio Emilia), Rita Cucchiara (University of Modena and Reggio Emilia)
ClassificationAnomaly DetectionTransformerDiffusion modelContrastive LearningImageBenchmark
🎯 What it does: This study proposes CoDE, a contrastive learning-based deep fake embedding space, specifically designed to distinguish real images from forged images generated by diffusion models, and creates the D3 dataset containing 9.2M images.
Contrastive ground-level image and remote sensing pre-training improves representation learning for natural world imagery
Andy V Huynh, Moisés Expósito-Alonso (Stanford University)
RecognitionTransformerAuto EncoderContrastive LearningImageBenchmark
🎯 What it does: This paper proposes CRISP (ContRastive Image-remote Sensing Pre-training), a self-supervised multi-view contrastive learning framework, and constructs a large-scale Nature Multi-View (NMV) dataset for pre-training on natural world images (including ground and aerial perspectives) to enhance the performance of fine-grained species identification and related ecological tasks.
Contrastive Learning with Synthetic Positives
Dewen Zeng (University of Notre Dame), Yiyu Shi (University of Notre Dame)
Data SynthesisRepresentation LearningDiffusion modelContrastive LearningImage
🎯 What it does: Propose the CLSP method, which generates hard positive samples using an unsupervised diffusion model and incorporates them into a contrastive learning framework