ICCV 2023 Papers — Page 10
IEEE/CVF International Conference on Computer Vision · 2156 papers
ICD-Face: Intra-class Compactness Distillation for Face Recognition
Zhipeng Yu (University of Chinese Academy of Sciences), Ding Liang (SenseTime Research)
RecognitionKnowledge DistillationContrastive LearningImage
🎯 What it does: This paper proposes a knowledge distillation framework for lightweight face recognition, ICD-Face, which enhances the intra-class similarity of the student model by adding intra-class compact distillation on the basis of traditional feature consistency distillation.
ICE-NeRF: Interactive Color Editing of NeRFs via Decomposition-Aware Weight Optimization
Jae-Hyeok Lee (KAIST), Dae-Shik Kim (KAIST)
SegmentationOptimizationNeural Radiance FieldImage
🎯 What it does: The ICE-NeRF framework is proposed by fine-tuning the local weights of a pre-trained NeRF model and utilizing user-provided rough masks for interactive color editing.
ICICLE: Interpretable Class Incremental Continual Learning
Dawid Rymarczyk (Jagiellonian University), Bartlomiej Twardowski
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: An interpretable category incremental continual learning method called ICICLE is proposed, which gradually expands knowledge through prototype component learning.
ICL-D3IE: In-Context Learning with Diverse Demonstrations Updating for Document Information Extraction
Jiabang He (University of Electronic Science and Technology of China), Heng Tao Shen (University of Electronic Science and Technology of China)
Data-Centric LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes the ICL-D3IE framework for Document Information Extraction (DIE) based on large language models, utilizing a small number of examples to achieve zero-shot In-Context Learning.
iDAG: Invariant DAG Searching for Domain Generalization
Zenan Huang (Zhejiang University), Nenggan Zheng (Zhejiang University)
Domain AdaptationGraph Neural NetworkContrastive LearningImage
🎯 What it does: A learning framework based on immutable directed acyclic graphs (iDAG) is proposed, which utilizes a learnable adjacency matrix and prototype approximation to recover causal structures from multi-domain data, thereby achieving domain generalization.
Identification of Systematic Errors of Image Classifiers on Rare Subgroups
Jan Hendrik Metzen (Bosch Center for Artificial Intelligence), Dan Zhang (Bosch Center for Artificial Intelligence)
ClassificationAdversarial AttackConvolutional Neural NetworkPrompt EngineeringDiffusion modelImage
🎯 What it does: By using a text-to-image model to generate controllable images of rare subgroups and employing combinatorial testing for model inference, a method called PROMPTATTACK is proposed to identify systematic errors of image classifiers on rare subgroups.
Identity-Consistent Aggregation for Video Object Detection
Chaorui Deng (Australia Institute of Machine Learning), Qi Wu (Australia Institute of Machine Learning)
Object DetectionTransformerContrastive LearningVideo
🎯 What it does: We propose ClipVID, an end-to-end video object detection model that enhances object representation by utilizing identity-consistent temporal context aggregation and achieves clip-level parallel prediction.
Identity-Seeking Self-Supervised Representation Learning for Generalizable Person Re-Identification
Zhaopeng Dou (Tsinghua University), Shengjin Wang (Tsinghua University)
RecognitionRetrievalRepresentation LearningTransformerContrastive LearningImageVideo
🎯 What it does: An unsupervised, unlabelled identity-seeking self-supervised representation learning (ISR) method is proposed, which learns a human re-identification model that can efficiently perform in unseen domains using large-scale video data.
IDiff-Face: Synthetic-based Face Recognition through Fizzy Identity-Conditioned Diffusion Model
Fadi Boutros (Fraunhofer Institute for Computer Graphics Research), Naser Damer (Fraunhofer Institute for Computer Graphics Research)
RecognitionGenerationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes a method for generating synthetic facial data based on an identity-conditioned diffusion model called IDiff-Face, which can provide high-quality training samples with identity distinguishability and intra-class diversity for facial recognition models.
IHNet: Iterative Hierarchical Network Guided by High-Resolution Estimated Information for Scene Flow Estimation
Yun Wang (Huazhong University of Science and Technology), Xin Yang (Huazhong University of Science and Technology)
Depth EstimationAutonomous DrivingConvolutional Neural NetworkOptical FlowPoint Cloud
🎯 What it does: This paper proposes an Iterative Hierarchical Structure-based Scene Flow Estimation Network (IHNet), which improves motion estimation between point clouds by guiding the current level with the high-resolution estimation results from the previous iteration.
IIEU: Rethinking Neural Feature Activation from Decision-Making
Sudong Cai (Kyoto University)
ClassificationObject DetectionComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A neural feature activation model IIEU based on a multi-criteria decision-making (MCDM) perspective is proposed to address the feature scoring mismatch problem and enhance the performance of ReLU and SOTA activation functions.
Image-Free Classifier Injection for Zero-Shot Classification
Anders Christensen (Technical University of Denmark), Zeynep Akata (Max Planck Institute for Intelligent Systems)
ClassificationRepresentation LearningAuto EncoderContrastive LearningImageText
🎯 What it does: Injecting a classifier for new categories into a pre-trained classification model to achieve zero-shot classification without any image data.
ImbSAM: A Closer Look at Sharpness-Aware Minimization in Class-Imbalanced Recognition
Yixuan Zhou (University of Electronic Science and Technology of China), Hengtao Shen (Peng Cheng Laboratory)
ClassificationRecognitionAnomaly DetectionOptimizationSupervised Fine-TuningImage
🎯 What it does: In class-imbalanced recognition tasks, this paper conducts an in-depth study of the overfitting bottleneck from a generalization perspective and proposes an improved Sharpness-Aware Minimization (ImbSAM) method for tail classes.
ImGeoNet: Image-induced Geometry-aware Voxel Representation for Multi-view 3D Object Detection
Tao Tu (National Tsing Hua University), Min Sun (Amazon)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a 3D object detection framework called ImGeoNet based on multi-view images, utilizing image-induced geometric perception voxel representation for detection.
Imitator: Personalized Speech-driven 3D Facial Animation
Balamurugan Thambiraja (Max Planck Institute for Intelligent Systems), Justus Thies (Max Planck Institute for Intelligent Systems)
GenerationTransformerVideoAudio
🎯 What it does: Extracting speaker style through short videos, utilizing a general Transformer to generate voice-driven 3D facial animations, outputting personalized and well-synchronized lip animations.
Implicit Autoencoder for Point-Cloud Self-Supervised Representation Learning
Siming Yan (University of Texas at Austin), Qixing Huang (University of Texas at Austin)
Object DetectionSegmentationRepresentation LearningGraph Neural NetworkTransformerAuto EncoderPoint Cloud
🎯 What it does: This paper proposes an Implicit AutoEncoder (IAE) that uses implicit surface representation to address the sampling variation problem in self-supervised representation learning of point clouds by changing the decoder output to an implicit function.
Implicit Identity Representation Conditioned Memory Compensation Network for Talking Head video Generation
Fa-Ting Hong (Hong Kong University of Science and Technology), Dan Xu (Hong Kong University of Science and Technology)
GenerationData SynthesisConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: A MCNet network based on a global facial meta-memory bank and implicit identity representation is proposed, utilizing a memory compensation mechanism to enhance the quality of speaker video generation from single-frame source images under significant motion/occlusion.
Implicit Neural Representation for Cooperative Low-light Image Enhancement
Shuzhou Yang (Peking University), Jian Zhang (Peking University)
RestorationGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A framework for low-light image enhancement using implicit neural representation (NeRCo) is proposed, capable of recovering visually pleasing high-brightness images under unsupervised conditions.
Implicit Temporal Modeling with Learnable Alignment for Video Recognition
Shuyuan Tu (Fudan University), Yu-Gang Jiang (Fudan University)
RecognitionComputational EfficiencyTransformerContrastive LearningVideo
🎯 What it does: An Implicit Learnable Alignment (ILA) method is proposed, which uses an implicit learnable mask for coarse alignment of adjacent frames, replacing traditional spatiotemporal attention for temporal modeling in video recognition.
Improved Knowledge Transfer for Semi-Supervised Domain Adaptation via Trico Training Strategy
Ba Hung Ngo (Dongguk University), Sung In Cho (Dongguk University)
Domain AdaptationGraph Neural NetworkImage
🎯 What it does: This paper proposes a semi-supervised domain adaptation method called TriCT, which utilizes a multi-layer perceptron (MLP) and two graph convolutional networks (inter-view GCN and intra-view GCN) through three collaborative training strategies to address cross-domain and intra-domain differences.
Improved Visual Fine-tuning with Natural Language Supervision
Junyang Wang (Beijing Jiaotong University), Qi Qian (Alibaba Group)
ClassificationRecognitionTransformerSupervised Fine-TuningContrastive LearningImageText
🎯 What it does: This paper proposes a natural language supervision method based on a fixed text encoder (TeS), which regularizes the classifier of the visual model by introducing a text reference distribution, alleviating the conflict between pre-trained model bias and catastrophic forgetting.
Improving 3D Imaging with Pre-Trained Perpendicular 2D Diffusion Models
Suhyeon Lee (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
RestorationGenerationDiffusion modelImageBiomedical DataMagnetic Resonance ImagingComputed TomographyStochastic Differential Equation
🎯 What it does: Using two pre-trained 2D diffusion models in the vertical direction to solve 3D inverse problems, constructing generation and reconstruction of 3D volumes.
Improving Adversarial Robustness of Masked Autoencoders via Test-time Frequency-domain Prompting
Qidong Huang (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
ClassificationAdversarial AttackAuto EncoderImage
🎯 What it does: This paper studies the robustness differences of the visual BERT pre-training method (especially MAE) under adversarial attacks and proposes a scheme to enhance the robustness of MAE during the testing phase through frequency domain visual prompts.
Improving CLIP Fine-tuning Performance
Yixuan Wei (Tsinghua University), Baining Guo (Microsoft Research Asia)
Object DetectionSegmentationDepth EstimationKnowledge DistillationTransformerContrastive LearningImage
🎯 What it does: This paper distills the pre-trained CLIP model using a classic feature map distillation framework, significantly improving its fine-tuning performance on downstream tasks while maintaining its original semantic expressiveness. The framework is also extended to models such as DINO, DeiT, and SwinV2-G.
Improving Continuous Sign Language Recognition with Cross-Lingual Signs
Fangyun Wei (Microsoft Research Asia), Yutong Chen (Microsoft Research Asia)
RecognitionVideo
🎯 What it does: This paper enhances the training of monolingual continuous sign language recognition models by identifying cross-linguistic gestures (visually similar gestures across different sign languages) and utilizing multilingual datasets.
Improving Diversity in Zero-Shot GAN Adaptation with Semantic Variations
Seogkyu Jeon (Yonsei University), Hyeran Byun (Yonsei University)
GenerationDomain AdaptationPrompt EngineeringGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a zero-shot GAN adaptation method that guides a pre-trained StyleGAN to generate target domain images through CLIP text prompts, and enhances the diversity of the generated images based on this.
Improving Equivariance in State-of-the-Art Supervised Depth and Normal Predictors
Yuanyi Zhong (University of Illinois Urbana-Champaign), David Forsyth (University of Illinois Urbana-Champaign)
Depth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper studies the lack of invariance in depth and surface normal prediction models under cropping-scaling transformations, and proposes a training framework based on invariance regularization that can enhance the model's invariance and accuracy without increasing inference costs.
Improving Generalization in Visual Reinforcement Learning via Conflict-aware Gradient Agreement Augmentation
Siao Liu (Fudan University), Zhongxue Gan (Fudan University)
Robotic IntelligenceReinforcement LearningVideoBenchmark
🎯 What it does: A conflict-aware gradient consistency enhancement framework (CG2A) is proposed in visual reinforcement learning, which enhances the generalization ability of the policy by combining various data augmentations and utilizing a gradient consistency solver and soft gradient surgery.
Improving Generalization of Adversarial Training via Robust Critical Fine-Tuning
Kaijie Zhu (University of Chinese Academy of Sciences), Ge Yang (Chinese Academy of Sciences)
OptimizationAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: On existing adversarial training models, only the least important modules are fine-tuned, and interpolation is used to enhance generalization ability.
Improving Lens Flare Removal with General-Purpose Pipeline and Multiple Light Sources Recovery
Yuyan Zhou (Nanjing University of Aeronautics and Astronautics), Chongyi Li (Nankai University)
RestorationData SynthesisTransformerImage
🎯 What it does: This paper proposes a new pixel-level convex combination synthesis process based on ISP to generate more realistic halo pollution images, and designs a multi-source recovery strategy that does not require hard thresholds. It also contributes a real halo test dataset across multiple devices; the U-Former/UNet model trained based on this scheme performs better on various benchmarks.
Improving Online Lane Graph Extraction by Object-Lane Clustering
Yigit Baran Can (ETH Zurich), Luc Van Gool (ETH Zurich)
Object DetectionSegmentationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: By associating 3D object detection results with lane centerlines, an object-lane clustering loss is proposed to improve the extraction of online BEV lane maps.
Improving Pixel-based MIM by Reducing Wasted Modeling Capability
Yuan Liu (Shanghai AI Laboratory), Dahua Lin (Shanghai AI Laboratory)
Object DetectionSegmentationRepresentation LearningTransformerSupervised Fine-TuningImage
🎯 What it does: This study investigates the low-frequency feature bias of pixel-level MIM and proposes a multi-layer feature fusion method to enhance the representation capability of MAE and PixMIM.
Improving Representation Learning for Histopathologic Images with Cluster Constraints
Weiyi Wu (Dartmouth College), Saeed Hassanpour (Dartmouth College)
Representation LearningConvolutional Neural NetworkContrastive LearningImageBiomedical Data
🎯 What it does: A self-supervised learning framework called CluSiam is proposed, which combines clustering constraints for representation learning and clustering of unlabeled pathological images.
Improving Sample Quality of Diffusion Models Using Self-Attention Guidance
Susung Hong (Korea University), Seungryong Kim (Korea University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a Self-Attention Guidance (SAG) technique that utilizes internal self-attention maps to adversarially blur focus areas during the reverse sampling process of diffusion models, further guiding the model to generate higher quality images with fewer artifacts. Additionally, a simple Blur Guidance is introduced as the foundation of SAG.
Improving Transformer-based Image Matching by Cascaded Capturing Spatially Informative Keypoints
Chenjie Cao (Fudan University), Yanwei Fu (Fudan University)
Pose EstimationRetrievalTransformerImage
🎯 What it does: A cascade matching model based on Transformer, CasMTR, is proposed, achieving denser and more accurate image matching, supporting high-resolution images.
Improving Unsupervised Visual Program Inference with Code Rewriting Families
Aditya Ganeshan (Brown University), Daniel Ritchie (Brown University)
TransformerPoint Cloud
🎯 What it does: The Sparse Intermittent Rewrite Injection (SIRI) method is proposed to enhance program quality and training convergence speed in unsupervised visual program reasoning using a family of code rewrites.
In-Style: Bridging Text and Uncurated Videos with Style Transfer for Text-Video Retrieval
Nina Shvetsova (Goethe University Frankfurt), Hilde Kuehne (University of Bonn)
RetrievalContrastive LearningVideoText
🎯 What it does: A method for unannotated video retrieval that uses only text queries during the training phase, without corresponding videos, is proposed. It utilizes unorganized videos and style transfer to generate pseudo-labels and trains a dual-encoder for text-video retrieval.
Incremental Generalized Category Discovery
Bingchen Zhao (University of Edinburgh), Oisin Mac Aodha (University of Edinburgh)
ClassificationRecognitionSpiking Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: This paper proposes a method for Incremental General Category Discovery (IGCD) tasks that utilizes a non-parametric SNN classifier and density peak-based sample selection, balancing the classification of old and new categories while discovering new categories, and constructs the iNatIGCD benchmark dataset.
Indoor Depth Recovery Based on Deep Unfolding with Non-Local Prior
Yuhui Dai (East China Normal University), Guixu Zhang (East China Normal University)
RestorationDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: Utilizing deep unfolding networks combined with non-local autoregressive regularization to achieve the recovery of sparse depth maps to complete depth maps.
Inducing Neural Collapse to a Fixed Hierarchy-Aware Frame for Reducing Mistake Severity
Tong Liang (Ohio State University), Jim Davis (Ohio State University)
ClassificationOptimizationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A hierarchical perception framework (HAFrame) is proposed, which fixes this framework in the linear classifier of deep networks in advance, and uses cosine similarity auxiliary loss to cluster features onto the corresponding classifier vectors, thereby reducing the severity of errors.
InfiniCity: Infinite-Scale City Synthesis
Chieh Hubert Lin (University of California Merced), Sergey Tulyakov
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A three-stage pipeline is proposed, first generating city 2D maps of arbitrary scales through infinite pixel satellite image synthesis, then elevating the 2D maps to a 3D closed voxel world using octree-based voxel completion, and finally generating a navigable and realistic 3D urban environment through voxel-based neural rendering.
Informative Data Mining for One-Shot Cross-Domain Semantic Segmentation
Yuxi Wang (University of Chinese Academy of Sciences), Zhaoxiang Zhang (University of Chinese Academy of Sciences)
SegmentationDomain AdaptationContrastive LearningImage
🎯 What it does: In the scenario of single-image target domain adaptation, a framework named Informative Data Mining (IDM) is proposed, which can quickly adapt a pre-trained source domain semantic segmentation model using only one unlabeled target image and a limited number of iterations.
Inherent Redundancy in Spiking Neural Networks
Man Yao (Xi'an Jiaotong University), Guoqi Li (Chinese Academy of Sciences)
Spiking Neural NetworkVideo
🎯 What it does: This paper conducts a systematic analysis of the redundancy problem in Spiking Neural Networks (SNN) and proposes a high-level spatial attention (ASA) module that optimizes membrane potential distribution to reduce redundant spikes and enhance performance.
Innovating Real Fisheye Image Correction with Dual Diffusion Architecture
Shangrong Yang (Beijing Jiaotong University), Yao Zhao (Beijing Jiaotong University)
RestorationGenerationDiffusion modelImage
🎯 What it does: A Dual Diffusion Architecture (DDA) is proposed to achieve high-quality correction of real fisheye images.
Inspecting the Geographical Representativeness of Images from Text-to-Image Models
Abhipsa Basu (Indian Institute of Science), Danish Pruthi (Indian Institute of Science)
GenerationDiffusion modelImage
🎯 What it does: By comparing the geographic representativeness of images generated by DALL·E 2 and Stable Diffusion through human evaluation across different countries, this study analyzes their representativeness, authenticity, and the feasibility of automatic evaluation methods.
INSTA-BNN: Binary Neural Network with INSTAnce-aware Threshold
Changhun Lee (Pohang University of Science and Technology), Jae-Joon Kim (Seoul National University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: A binary neural network (BNN) architecture called INSTA-BNN is proposed, which dynamically adjusts thresholds using higher-order statistics of input instances to improve the accuracy of binarized networks.
Instance and Category Supervision are Alternate Learners for Continual Learning
Xudong Tian (East China Normal University), Yuan Xie (Contemporary Amperex Technology Co., Limited)
ClassificationRepresentation LearningConvolutional Neural NetworkSupervised Fine-TuningContrastive LearningImage
🎯 What it does: An alternating learning framework is proposed, which alternates between self-supervised learning (SSL) and supervised learning (SL), utilizing both instance-level and category-level supervision in continual learning.
Instance Neural Radiance Field
Yichen Liu (Hong Kong University of Science and Technology), Chi-Keung Tang (Hong Kong University of Science and Technology)
Object DetectionSegmentationNeural Radiance FieldPoint Cloud
🎯 What it does: Instance-NeRF is proposed, which adds an instance field on top of the pre-trained NeRF to achieve instance segmentation of unlabelled data in 3D space and generate continuous instance masks from any viewpoint.
Instance-aware Dynamic Prompt Tuning for Pre-trained Point Cloud Models
Yaohua Zha (Tsinghua University), Shu-Tao Xia (Harbin Institute of Technology)
RecognitionSegmentationTransformerPrompt EngineeringPoint Cloud
🎯 What it does: This paper implements parameter-efficient prompt tuning on a pre-trained point cloud model and proposes the Instance-aware Dynamic Prompt Tuning (IDPT) method.
Instruct-NeRF2NeRF: Editing 3D Scenes with Instructions
Ayaan Haque (University of California Berkeley), Angjoo Kanazawa (University of California Berkeley)
GenerationData SynthesisDiffusion modelNeural Radiance FieldImage
🎯 What it does: This paper proposes a method for consistent 3D editing of captured NeRF scenes using text instructions. The core idea is to iteratively update training images and edit them with InstructPix2Pix, ultimately achieving a 3D consistent scene that meets the instructions.
INT2: Interactive Trajectory Prediction at Intersections
Zhijie Yan (Beihang University), Hao Zhao (Beihang University)
Domain AdaptationAutonomous DrivingTime SeriesBenchmark
🎯 What it does: A large-scale interactive trajectory prediction dataset, INT2, is proposed, and the M2I model is benchmarked on this dataset.
Integrally Migrating Pre-trained Transformer Encoder-decoders for Visual Object Detection
Feng Liu (University of Chinese Academy of Sciences), Qixiang Ye (University of Chinese Academy of Sciences)
Object DetectionTransformerImage
🎯 What it does: Proposes to completely transfer the pre-trained Vision Transformer (ViT) encoder-decoder to the object detector, removing the traditional FPN and using only the decoder as the detection head, while adding a Multi-Scale Feature Modulator (MFM) to construct a 'fully pre-trained' feature extraction path.
Integrating Boxes and Masks: A Multi-Object Framework for Unified Visual Tracking and Segmentation
Yuanyou Xu (Zhejiang University), Yi Yang (Zhejiang University)
Object DetectionObject TrackingSegmentationConvolutional Neural NetworkTransformerVideo
🎯 What it does: A multi-object box-mask unified tracking and segmentation framework MITS is proposed, which uses a unified recognition module to support box and mask initialization, and introduces a pointer box predictor for precise box prediction.
IntentQA: Context-aware Video Intent Reasoning
Jiapeng Li (Xi'an Jiaotong University), Lifeng Fan (Beijing Institute for General Artificial Intelligence)
RetrievalRecommendation SystemTransformerPrompt EngineeringContrastive LearningVideoMultimodality
🎯 What it does: A new video question answering task, IntentQA, is proposed, focusing on video intent reasoning, and a corresponding large-scale dataset is constructed.
Inter-Realization Channels: Unsupervised Anomaly Detection Beyond One-Class Classification
Declan McIntosh (University of Victoria), Alexandra Branzan Albu (University of Victoria)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a completely unsupervised image anomaly detection and localization method called InReaCh, which constructs channels by associating similar patches between different training images, filters high-confidence normal patches to build a normal model, and then uses distance to discriminate new image patches for anomaly detection.
Interaction-aware Joint Attention Estimation Using People Attributes
Chihiro Nakatani (Toyota Technological Institute), Norimichi Ukita (Toyota Technological Institute)
RecognitionObject DetectionTransformerImageVideo
🎯 What it does: A joint attention estimation method based on Transformer is proposed, utilizing attributes such as face position, gaze direction, and actions, and achieving the prediction of common attention points for multiple people in a single image through interactive modeling.
Interactive Class-Agnostic Object Counting
Yifeng Huang (Stony Brook University), Minh Hoai (VinAI Research)
Object DetectionSegmentationImage
🎯 What it does: An interactive category-free object counting framework is proposed, allowing users to correct counting errors by clicking on areas and providing a range of numbers.
InterDiff: Generating 3D Human-Object Interactions with Physics-Informed Diffusion
Sirui Xu (University of Illinois at Urbana-Champaign), Liang-Yan Gui (University of Illinois at Urbana-Champaign)
GenerationPose EstimationTransformerDiffusion modelPoint CloudMesh
🎯 What it does: This paper proposes InterDiff, which utilizes diffusion models to predict 3D human-object interaction sequences and enhances physical feasibility through a physics-inspired corrector.
InterFormer: Real-time Interactive Image Segmentation
You Huang (Xiamen University), Rongrong Ji (Contemporary Amperex Technology Co. Limited)
SegmentationTransformerImage
🎯 What it does: This paper proposes InterFormer, which separates image preprocessing from interaction, using large-scale Vision Transformer for offline encoding, and employs a lightweight I-MSA module for real-time segmentation on CPU during interaction.
IntrinsicNeRF: Learning Intrinsic Neural Radiance Fields for Editable Novel View Synthesis
Weicai Ye (Zhejiang University), Guofeng Zhang (Zhejiang University)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: In a static scene with known camera poses from multiple perspectives, the invisible reflectance, shadows, and residual layers are decomposed in an editable manner, supporting online visual editing and new perspective synthesis with variable lighting.
Introducing Language Guidance in Prompt-based Continual Learning
Muhammad Gul Zain Ali Khan (RPTU), Muhammad Zeshan Afzal (DFKI)
ClassificationRepresentation LearningTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: This paper proposes a language-guided prompt learning method (LGCL) that enhances continual learning performance without replay buffers by constraining the keys and values of the prompt pool at both task and category levels using a pre-trained language encoder.
Invariant Feature Regularization for Fair Face Recognition
Jiali Ma (Panasonic), Hanwang Zhang (Nanyang Technological University)
RecognitionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a self-supervised invariant feature regularization (INV-REG) method that iteratively learns data partitioning and feature regularization to eliminate biases caused by representations such as race and gender, thereby enhancing fairness in facial recognition.
Invariant Training 2D-3D Joint Hard Samples for Few-Shot Point Cloud Recognition
Xuanyu Yi (Nanyang Technological University), Hanwang Zhang (Nanyang Technological University)
ClassificationRecognitionRetrievalGraph Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: The INVJOINT invariant training framework is proposed, which significantly improves few-shot point cloud classification performance by selecting and training joint hard samples between 2D and 3D models.
Inverse Compositional Learning for Weakly-supervised Relation Grounding
Huan Li (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
RecognitionObject DetectionTransformerImageVideo
🎯 What it does: A weakly supervised video relationship localization method is proposed, achieving alignment of relational semantics and visual features through holistic and local dual reasoning.
Inverse Problem Regularization with Hierarchical Variational Autoencoders
Jean Prost (University of Bordeaux), Nicolas Papadakis (University of Bordeaux)
RestorationSuper ResolutionAuto EncoderImage
🎯 What it does: A hierarchical variational autoencoder (HVAE) is proposed as a prior to solve linear inverse problems (deblurring, super-resolution, inpainting, etc.) by alternately optimizing images and latent variables.
IOMatch: Simplifying Open-Set Semi-Supervised Learning with Joint Inliers and Outliers Utilization
Zekun Li (Nanjing University), Yang Gao (Nanjing University)
ClassificationAnomaly DetectionContrastive LearningImage
🎯 What it does: This paper proposes an open semi-supervised learning framework named IOMatch, which can jointly utilize both internal and external samples from unlabeled data for training.
Is Imitation All You Need? Generalized Decision-Making with Dual-Phase Training
Yao Wei (Zhejiang University), Shuang Ma (Microsoft)
Robotic IntelligenceTransformerReinforcement LearningPrompt EngineeringSequential
🎯 What it does: We propose DualMind, a general decision-making agent that uses dual-stage training to directly execute tasks across multiple domains, scenarios, and different executors based on prompts.
Isomer: Isomerous Transformer for Zero-shot Video Object Segmentation
Yichen Yuan (Dalian University of Technology), Lei Zhang (Hong Kong Polytechnic University)
Object DetectionSegmentationTransformerOptical FlowVideo
🎯 What it does: A hierarchical heterogeneous Transformer (Isomer) is proposed for unsupervised video object segmentation.
IST-Net: Prior-Free Category-Level Pose Estimation with Implicit Space Transformation
Jianhui Liu (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
Pose EstimationConvolutional Neural NetworkPoint Cloud
🎯 What it does: An IST-Net without category priors is proposed, which maps camera space features to world space through an implicit space transformation module, directly achieving category-level 6D pose estimation.
Iterative Denoiser and Noise Estimator for Self-Supervised Image Denoising
Yunhao Zou (Beijing Institute of Technology), Ying Fu (Beijing Institute of Technology)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an iterative self-supervised denoising pipeline called DCD-Net, which first uses a blind spot network for rough denoising, and then generates noise pairs for N2N training through noise estimation, gradually approaching strong baseline performance from a single noisy image.
Iterative Prompt Learning for Unsupervised Backlit Image Enhancement
Zhexin Liang (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
RestorationPrompt EngineeringVision Language ModelImage
🎯 What it does: An unsupervised backlight image enhancement method is proposed by utilizing the prior knowledge of the CLIP vision-language model, through learnable prompts and iterative optimization;
Iterative Soft Shrinkage Learning for Efficient Image Super-Resolution
Jiamian Wang (Rochester Institute of Technology), Zhiqiang Tao (Rochester Institute of Technology)
RestorationSuper ResolutionOptimizationImage
🎯 What it does: An efficient single image super-resolution framework based on unstructured pruning is proposed, and an Iterative Soft Shrinkage Percentage (ISS-P) method is designed to directly train sparse networks from random initialization.
Iterative Superquadric Recomposition of 3D Objects from Multiple Views
Stephan Alaniz (University of Tübingen), Zeynep Akata (University of Tübingen)
SegmentationGenerationOptimizationImagePoint Cloud
🎯 What it does: Using multi-view 2D images, the target object is gradually reconstructed in 3D space using superquadrics to generate interpretable part abstractions.
ITI-GEN: Inclusive Text-to-Image Generation
Cheng Zhang (Carnegie Mellon University), Fernando De la Torre (Carnegie Mellon University)
GenerationPrompt EngineeringDiffusion modelImage
🎯 What it does: This study explores a method for learning pluggable prompt tokens using a small number of reference images, achieving fairness and multi-attribute consistency in text-to-image generation.
iVS-Net: Learning Human View Synthesis from Internet Videos
Junting Dong (Zhejiang University), Sida Peng (NetEase Games AI Lab)
GenerationData SynthesisPose EstimationNeural Radiance FieldGenerative Adversarial NetworkImageVideo
🎯 What it does: This study investigates a method for 3D human body model reconstruction and free-viewpoint synthesis based on a single image, utilizing monocular videos from the internet for self-supervised learning.
Joint Demosaicing and Deghosting of Time-Varying Exposures for Single-Shot HDR Imaging
Jungwoo Kim (KAIST), Min H. Kim (KAIST)
RestorationTransformerImageVideo
🎯 What it does: A single-shot HDR image reconstruction method for quad-Bayer (Quad-Bayer) time-varying exposure sensors is proposed, jointly addressing deblocking, ghosting removal, and deblurring to generate high-quality HDR images.
Joint Implicit Neural Representation for High-fidelity and Compact Vector Fonts
Chia-Hao Chen (Tsinghua University), Song-Hai Zhang (Tsinghua University)
GenerationCompressionImage
🎯 What it does: A joint implicit neural representation (SDF + corner point field) is proposed, achieving high-quality vector font generation through dual contours and curve fitting.
Joint Metrics Matter: A Better Standard for Trajectory Forecasting
Erica Weng (Carnegie Mellon University), Kris Kitani (Carnegie Mellon University)
TransformerAuto EncoderTime Series
🎯 What it does: This paper proposes the use of joint metrics (JADE/JFDE/collision rate) to evaluate trajectory prediction models, and based on this, improves the loss function to enhance the realism and interactivity of multi-agent predictions.
Joint-Relation Transformer for Multi-Person Motion Prediction
Qingyao Xu (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
Pose EstimationTransformerVideo
🎯 What it does: A dual-stream Joint-Relation Transformer is proposed for multi-person motion prediction, capable of simultaneously modeling skeletal joint information and inter-joint relationships, incorporating relationship-aware attention and relationship supervision into the Transformer.
JOTR: 3D Joint Contrastive Learning with Transformers for Occluded Human Mesh Recovery
Jiahao Li (Zhejiang University), Yi Yang (Zhejiang University)
Pose EstimationTransformerContrastive LearningMesh
🎯 What it does: This study focuses on 3D human mesh recovery from a single image in occluded scenes, proposing the JOTR framework that integrates 2D global features and 3D local features through a Transformer for coarse-to-fine alignment.
Jumping through Local Minima: Quantization in the Loss Landscape of Vision Transformers
Natalia Frumkin (University of Texas at Austin), Diana Marculescu (University of Texas at Austin)
CompressionOptimizationTransformerContrastive LearningImage
🎯 What it does: This paper proposes Evol-Q, a post-training quantization method for visual Transformers, which utilizes evolutionary search to make subtle perturbations on the quantization scale of each attention block and evaluates performance using infoNCE contrastive loss.
KECOR: Kernel Coding Rate Maximization for Active 3D Object Detection
Yadan Luo (University of Queensland), Zi Huang (University of Queensland)
Object DetectionAutonomous DrivingPoint Cloud
🎯 What it does: This paper proposes an active learning framework based on Kernel Encoding Rate Maximization (KECOR) for efficiently selecting the most informative point clouds for labeling in LiDAR 3D object detection tasks.
Keep It SimPool: Who Said Supervised Transformers Suffer from Attention Deficit?
Bill Psomas (National Technical University of Athens), Yannis Avrithis (Institute of Advanced Research in Artificial Intelligence)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: A simple attention-based pooling method called SimPool is proposed to replace the default pooling in convolutional and Transformer networks.
Kick Back & Relax: Learning to Reconstruct the World by Watching SlowTV
Jaime Spencer (University of Surrey), Richard Bowden (University of Surrey)
Depth EstimationDomain AdaptationConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: A self-supervised monocular depth estimation model is proposed, trained using a large-scale SlowTV video dataset extracted from YouTube, achieving zero-shot cross-domain generalization.
Knowing Where to Focus: Event-aware Transformer for Video Grounding
Jinhyun Jang (Yonsei University), Kwanghoon Sohn (Korea Institute of Science and Technology)
RecognitionSegmentationTransformerVideo
🎯 What it does: This paper proposes an event-aware video alignment framework EaTR, which utilizes event reasoning and moment reasoning to dynamically generate moment queries, achieving end-to-end video moment localization.
Knowledge Proxy Intervention for Deconfounded Video Question Answering
Jiangtong Li (Shanghai Jiao Tong University), Liqing Zhang (Shanghai Jiao Tong University)
RecognitionRetrievalVideo
🎯 What it does: A model-free knowledge proxy intervention (KPI) framework is proposed, utilizing front-door adjustment to eliminate dataset bias in video question answering;
Knowledge Restore and Transfer for Multi-Label Class-Incremental Learning
Songlin Dong (Xi'an Jiaotong University), Yihong Gong (Huawei Technologies)
ClassificationKnowledge DistillationTransformerImage
🎯 What it does: An end-to-end framework for multi-label category incremental learning (MLCIL) called KRT is proposed, which recovers old class knowledge through dynamic pseudo-labels and utilizes incremental cross-attention to transfer old knowledge.
Knowledge-Aware Federated Active Learning with Non-IID Data
Yu-Tong Cao (University of Sydney), Dacheng Tao (ShanghaiTech University)
Federated LearningKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A federated active learning framework KAFAL is designed for non-IID data, which efficiently learns a global model with a limited labeling budget by utilizing knowledge-specialized active sampling and knowledge compensation federated updates while preserving data privacy.
Knowledge-Aware Prompt Tuning for Generalizable Vision-Language Models
Baoshuo Kan (Qilu University of Technology), Feng Zheng (Southern University of Science and Technology)
ClassificationDomain AdaptationTransformerPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Proposes the Knowledge-Aware Prompt Tuning (KAPT) framework, which incorporates discrete and continuous prompts of external knowledge (from Wikipedia) into the CLIP vision-language model, and implements adaptive feature extraction for tasks with a visual adaptation head.
Knowledge-Spreader: Learning Semi-Supervised Facial Action Dynamics by Consistifying Knowledge Granularity
Xiaotian Li (State University of New York at Binghamton), Lijun Yin (State University of New York at Binghamton)
RecognitionKnowledge DistillationTransformerVideoMultimodality
🎯 What it does: A lightweight, online semi-supervised framework called Knowledge-Spreader is designed and implemented to learn dynamic facial action unit (AU) detection using sparse keyframe labels.
L-DAWA: Layer-wise Divergence Aware Weight Aggregation in Federated Self-Supervised Visual Representation Learning
Yasar Abbas Ur Rehman (TCL AI Lab), Nicholas D. Lane (University of Cambridge)
Federated LearningRepresentation LearningContrastive LearningImage
🎯 What it does: This paper proposes a hierarchical aggregation strategy based on angular differences for federated self-supervised visual representation learning, addressing the issues of model drift and bias caused by client heterogeneity.
LA-Net: Landmark-Aware Learning for Reliable Facial Expression Recognition under Label Noise
Zhiyu Wu (Peking University), Jinshi Cui (Peking University)
ClassificationRecognitionContrastive LearningImage
🎯 What it does: This paper proposes a Landmark-Aware Net (LA-Net) that enhances robustness in facial expression recognition tasks with label noise by utilizing facial keypoint information to generate sample label distributions and performing contrastive learning between expression features and keypoint features.
Label Shift Adapter for Test-Time Adaptation under Covariate and Label Shifts
Sunghyun Park (Qualcomm AI Research), Sungrack Yun (Qualcomm AI Research)
Domain AdaptationSupervised Fine-TuningImage
🎯 What it does: A pluggable Label Shift Adapter is proposed, which uses the label distribution as a condition to generate classifier parameters and normalization parameters, and achieves online adaptation under the resonance of covariate and label distribution by combining existing Test-Time Adaptation methods.
Label-Efficient Online Continual Object Detection in Streaming Video
Jay Zhangjie Wu (Show Lab), Mike Zheng Shou (Show Lab)
Object DetectionVideo
🎯 What it does: A more practical label-efficient online continuous object detection problem (LEOCOD) is proposed, along with a pluggable Efficient-CLS module to address it;
Label-Free Event-based Object Recognition via Joint Learning with Image Reconstruction from Events
Hoonhee Cho (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
RecognitionRestorationObject DetectionContrastive LearningImage
🎯 What it does: A joint learning framework is proposed that utilizes event camera data to simultaneously achieve object recognition and image reconstruction without using labels and paired images.
Label-Guided Knowledge Distillation for Continual Semantic Segmentation on 2D Images and 3D Point Clouds
Ze Yang (Nanyang Technological University), Guosheng Lin (Nanyang Technological University)
SegmentationKnowledge DistillationImagePoint Cloud
🎯 What it does: Proposes the Label-Guided Knowledge Distillation (LGKD) method, which utilizes the labels of the current step to guide the transfer of background probabilities from the old model, thereby addressing the novel-background conflict issue in continuous semantic segmentation, and validates its effectiveness on both 2D and 3D data.
Label-Noise Learning with Intrinsically Long-Tailed Data
Yang Lu (Xiamen University), Hanzi Wang (Xiamen University)
ClassificationData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a label noise learning framework called TABASCO for intrinsic long-tail distributions, enhancing model robustness through a two-stage dual-dimensional sample selection.
LAC - Latent Action Composition for Skeleton-based Action Segmentation
Di Yang (Inria), Francois Bremond (Inria)
SegmentationRepresentation LearningAuto EncoderContrastive LearningVideo
🎯 What it does: Using self-supervised action synthesis and contrastive learning, a transferable skeleton visual encoder is constructed for untrimmed skeleton action segmentation.
LAN-HDR: Luminance-based Alignment Network for High Dynamic Range Video Reconstruction
Haesoo Chung (Seoul National University), Nam Ik Cho (Seoul National University)
RestorationGenerationConvolutional Neural NetworkVideo
🎯 What it does: An end-to-end HDR video reconstruction framework called LAN-HDR is proposed, which aligns and completes details of low dynamic range (LDR) video frames using brightness information, and generates flicker-free HDR videos through temporal consistency loss.
Landscape Learning for Neural Network Inversion
Ruoshi Liu (Columbia University), Carl Vondrick (Columbia University)
GenerationPose EstimationOptimizationAdversarial AttackGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes a method to reshape the loss landscape in optimization-based inference by learning a mapping network, allowing gradient descent to be faster and more stable in a new space, applicable to tasks such as GAN inversion, 3D pose reconstruction, and adversarial defense.