CVPR 2026 Papers — Page 29

ReasonEdit: Towards Reasoning-Enhanced Image Editing Models

Fukun Yin (StepFun), Daxin Jiang (StepFun)

🎯 What it does: Proposed a novel image editing framework called ReasonEdit, which incorporates dual reasoning mechanisms of Thinking and Reflection, enabling dynamic parsing of abstract instructions and self-correction during the editing process.

Reasoning Diffusion for Unpaired Test Time Out-of-distribution Text-Image to Video Generation

Zirui Pan (Tsinghua University), Wenwu Zhu (Tsinghua University)

🎯 What it does: Propose the ReasonDiff model to address the text-image to video generation problem when no paired text and image inputs are available during testing, achieving cross-modal reasoning and alignment through two modules: VisionNarrator and AlignFormer.

Reasoning Palette: Modulating Reasoning via Latent Contextualization for Controllable Exploration for (V)LMs

Rujiao Long (Alibaba Group), Bo Zheng (Alibaba Group)

🎯 What it does: Propose the Reasoning Palette framework, which learns a sampleable latent context through VAE and injects the decoded prefix into the input of LLM/VLM to guide the model's internal reasoning path before generation; on this basis, perform brief SFT adaptation and schedule exploration and exploitation at the policy level during RL training.

Reasoning-Driven Anomaly Detection and Localization with Image-Level Supervision

Yizhou Jin (Beihang University), Yunhong Wang (Beihang University)

🎯 What it does: This paper proposes an end-to-end industrial anomaly detection framework based on a multimodal large language model (MLLM), which realizes anomaly detection, pixel-level localization, and interpretable reasoning through the model's own reasoning process, without relying on external visual modules or pixel-level annotations.

ReasonMap: Towards Fine-Grained Visual Reasoning from Transit Maps

Sicheng Feng (Westlake University), Xinchao Wang (National University of Singapore)

🎯 What it does: Developed a fine-grained visual reasoning benchmark called REASONMAP based on subway maps from 30 cities, along with 1,008 human-validated question-answer pairs.

ReasonX: MLLM-Guided Intrinsic Image Decomposition

Alara Dirik (Imperial College London), Anna Frühstück (Adobe Research)

🎯 What it does: Propose the ReasonX framework, which evaluates relative intrinsic image decomposition through multimodal large language models (MLLM) and performs fine-tuning on unlabeled real images using Group Relative Policy Optimization (GRPO).

ReAttnCLIP: Training-Free Open-Vocabulary Remote Sensing Image Segmentation via Re-defined Attention in CLIP

Xin Niu (Renmin University of China), Bing Su (Renmin University of China)

🎯 What it does: Propose a training-free open-vocabulary segmentation method for remote sensing images, ReAttnCLIP, which redefines the CLIP attention mechanism to achieve pixel-level high-quality feature extraction.

ReBaPL: Repulsive Bayesian Prompt Learning

Yassir Bendou (Sigma Nova), Mike Gartrell (Sigma Nova)

🎯 What it does: Proposes the Repulsive Bayesian Prompt Learning (ReBaPL) framework, which significantly enhances the generalization ability of few-shot vision-language models by introducing Bayesian inference and cyclic stochastic gradient Hamiltonian Monte Carlo (rcSGHMC) to sample the posterior distribution of prompts.

RebRL: Reinforcing Discrete Visual Diffusion Models with Rebalanced Timestep Credits

Mu Zhang (University of Chinese Academy of Sciences), Qixiang Ye (University of Chinese Academy of Sciences)

🎯 What it does: Propose RebRL in discrete diffusion models, improving gradient imbalance issues in RL training through rebalancing time-step credit allocation

ReCALL: Recalibrating Capability Degradation for MLLM-based Composed Image Retrieval

Tianyu Yang (Chinese Academy of Sciences), Tat-Seng Chua (National University of Singapore)

🎯 What it does: Proposes a diagnosis-generation-correction framework named ReCALL to address the capability degradation issue when migrating generative multi-modal large language models (MLLM) to retrieval models;

RecEdit-Drive: 3D Reconstruction-Guided Spatiotemporal Video Editing for Autonomous Driving Scenes

Yipeng Wu (Tianjin University), Di Lin (Tianjin University)

🎯 What it does: Propose the RecEdit-Drive framework for achieving high-quality, spatiotemporally consistent foreground object editing (deletion, replacement, insertion, relocation) in autonomous driving scenarios

Reclaiming Lost Text Layers for Source-Free Cross-Domain Few-Shot Learning

Zhenyu Zhang (Huazhong University of Science and Technology), Ruixuan Li (Huazhong University of Science and Technology)

🎯 What it does: This paper investigates the phenomenon of 'lost layers' in the CLIP text encoder within the source-free cross-domain few-shot learning (SFCDFSL) task, where certain intermediate layers are considered redundant but actually contribute to performance. It proposes a model named VtT, which reutilizes these lost layers through hierarchical fusion, text information absorption, and dynamic gradient supervision to enhance model performance in cross-domain few-shot tasks.

ReCoFuse: Ultra-Robust Image Fusion via Restorative Multi-Modal Diffusion Reciprocal Coupling

Hao Zhang (Wuhan University), Jiayi Ma (Wuhan University)

🎯 What it does: Proposes a super-robust image fusion framework named ReCoFuse based on restorative multimodal diffusion mutual coupling, which can automatically restore and fuse visible and infrared images under complex distortion conditions such as low light, haze, noise, low contrast, and stripes.

Reconstructing CLIP for Open-Vocabulary Dense Perception

Yajie Liu (Beihang University), Di Huang (Beihang University)

🎯 What it does: Propose the DenseRC framework, which reconstructs CLIP's dense representations using multi-layer value embeddings to achieve open-vocabulary dense perception (e.g., object detection, semantic segmentation).

Reconstructing Spiking Neural Networks Using a Single Neuron with Autapses

Wuque Cai (University of Electronic Science and Technology of China), Daqing Guo (University of Electronic Science and Technology of China)

🎯 What it does: This paper proposes a Time-Delayed Self-Synaptic (TDA) Spiking Neural Network (TDA-SNN), which achieves various SNN structures such as storage, feedforward, and convolutional architectures at the single-cell level by introducing self-synaptic delay feedback into a single Leaky Integrate-and-Fire (LIF) neuron.

Reconstruction-Guided Slot Curriculum: Addressing Object Over-Fragmentation in Video Object-Centric Learning

WonJun Moon (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)

🎯 What it does: Propose the SlotCurri method, which gradually expands video object slots through reconstruction-guided slot scheduling to address the over-fragmentation problem.

Recover to Predict: Progressive Retrospective Learning for Variable-Length Trajectory Prediction

Hao Zhou (Great Bay University), Fei Luo (Great Bay University)

🎯 What it does: Propose a Progressive Retrospective Framework (PRF) that gradually maps variable-length, missing trajectories to standard complete trajectories through multi-level recursive units, achieving high-precision prediction for incomplete observations.

Recovering Physically Plausible Human-Object Interactions from Monocular Videos

Dingbang Huang (University of Texas at Austin), Georgios Pavlakos (University of Texas at Austin)

🎯 What it does: Recover physically plausible full human-object interactions from monocular videos using reinforcement learning to correct noisy motion trajectories in physics simulations.

RecoverMark: Robust Watermarking for Localization and Recovery of Manipulated Faces

Haonan An (City University of Hong Kong), Yuguang Fang (City University of Hong Kong)

🎯 What it does: Designed RecoverMark, a robust watermarking framework that enables tamper localization, content recovery, and copyright verification in facial images.

RECS4R: Bridging Semantics and Geometry for Referring Remote Sensing Interpretation

Jinming Chai (Xidian University), Weibin Li (Xidian University)

🎯 What it does: This paper proposes RECS4R, a unified multi-task framework for visual grounding (VG) and referential semantic segmentation (RIS) in remote sensing images, achieving precise localization and segmentation through language-guided unified contour decoding, residual coarse-to-fine fusion, channel-isolated multi-scale fusion, and gradient consistency loss.

Rectifying Latent Space for Generative Single-Image Reflection Removal

Mingjia Li (Tianjin University), Xiaojie Guo (Tianjin University)

🎯 What it does: Proposes a single image reflection removal method based on an improved latent diffusion model, integrating reflection-equivalent VAE, learnable task-specific text embeddings, and depth-guided early branch sampling.

RecTok: Reconstruction Distillation along Rectified Flow

Qingyu Shi (Peking University), Xuelong Li (Nanyang Technological University)

🎯 What it does: Propose a high-dimensional visual tokenizer RecTok, and enhance semantics at each state of the forward flow through flow semantic distillation (FSD) and reconstruction alignment distillation (RAD), significantly improving reconstruction quality and generation performance.

Recurrent Reasoning with Vision-Language Models for Estimating Long-Horizon Embodied Task Progress

Yuelin Zhang (Renmin University of China), Wenbing Huang (Renmin University of China)

🎯 What it does: Propose a vision-language model R VLM based on iterative reasoning, which progressively estimates the progress of long-term tasks by utilizing local video clips and chain-of-thought (CoT).

Recurrent Video Masked Autoencoders

Daniel Zoran, Andrew Zisserman (Google DeepMind)

🎯 What it does: Developed a self-supervised video mask autoencoder (RVM) based on recursive Transformer, learning video representations through recursive aggregation and pixel reconstruction.

Red-teaming Retrieval-Augmented Diffusion Models via Poisoning Knowledge Bases

Xinqi Lyu (Hong Kong Polytechnic University), Bin Xiao (Hong Kong Polytechnic University)

🎯 What it does: A joint optimization backdoor attack was designed in black-box retrieval-augmented diffusion models, leveraging a small number of polluted knowledge base images and reinforcement learning to obtain trigger words, enabling协同 work between retrieval and generation stages.

ReDirector: Creating Any-Length Video Retakes with Rotary Camera Encoding

Byeongjun Park (EverEx), Jong Chul Ye (KAIST)

🎯 What it does: Proposed a video rephotography generation method called ReDirector, which can handle input videos of arbitrary length under dynamic camera motion and generate realistic videos that maintain multi-view consistency according to the target camera trajectory.

Reevaluating the Intra-Modal Misalignment Hypothesis in CLIP

Jonas Herzog (Zhejiang University), Yue Wang (Zhejiang University)

🎯 What it does: This paper re-evaluates the 'single-modal error' hypothesis within CLIP's 'cross-modal alignment' framework through both theoretical and empirical analyses, examining whether this hypothesis truly limits performance in image tasks.

Ref4D-VideoBench: Four-Dimensional Reference-Based Evaluation of Text-to-Video Generative Models

Jiajia Wei (Sun Yat-sen University), Weibin Wu (Sun Yat-sen University)

🎯 What it does: Propose a four-dimensional evaluation framework called Ref4D-VideoBench based on reference videos, which assesses semantic consistency, motion consistency, event temporal consistency, and world knowledge consistency of text-to-video generation models.

Refacade: Editing Object with Given Reference Texture

Youze Huang (University of Electronic Science and Technology of China), Rong Xiao (IntelliFusion Inc.)

🎯 What it does: Propose Refacade, addressing the object retexturing task in videos, which can precisely transfer textures from reference images to target objects while preserving the target's geometric structure.

ReFAct: Empowering Multimodal Web Agents with Visual and Context Focusing

Rui Wu (Ant Group), Yong Li (Ant Group)

🎯 What it does: Propose the ReFAct framework, endowing multi-modal Web agents with active visual grounding and memory focusing (defocus/refocus) capabilities, and train the ReFAct-7B model based on this framework;

RefAV: Towards Planning-Centric Scenario Mining

Cainan Davidson (Carnegie Mellon University), Neehar Peri (Carnegie Mellon University)

🎯 What it does: This paper proposes a new method for scene mining in autonomous driving scenarios—RefAV, which combines vision-language models (VLM) and program synthesis techniques to automatically identify and locate complex multi-agent interaction scenes in driving logs;

Refer-Agent: A Collaborative Multi-Agent System with Reasoning and Reflection for Referring Video Object Segmentation

Haichao Jiang (Sun Yat-sen University), Jian-Fang Hu (Sun Yat-sen University)

🎯 What it does: This paper proposes a zero-training, collaborative multi-agent system called Refer-Agent for video object segmentation based on text queries;

Refining Few-Step Text-to-Multiview Diffusion via Reinforcement Learning

Ziyi Zhang (Wuhan University), Lefei Zhang (Wuhan University)

🎯 What it does: Propose MVC-ZigAL, a reinforcement learning fine-tuning framework for few-step text-to-multi-view diffusion models;

Reflection Separation from a Single Image via Joint Latent Diffusion

Zheng-Hui Huang (Shanda AI Research Tokyo), Yung-Yu Chuang (National Taiwan University)

🎯 What it does: Proposed a single-image reflection separation method based on diffusion models, achieving reliable separation of images with strong or weak reflections by jointly generating transmission and reflection layers.

ReflexSplit: Single Image Reflection Separation via Layer Fusion-Separation

Chia-Ming Lee (National Yang Ming Chiao Tung University), Yu-Lun Liu (National Yang Ming Chiao Tung University)

🎯 What it does: This paper proposes a dual-stream framework called ReflexSplit for single image reflection separation, decomposing the mixed image into transmission and reflection layers.

ReFlow: Self-correction Motion Learning for Dynamic Scene Reconstruction

Yanzhe Liang (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)

🎯 What it does: ReFlow proposes a unified framework that self-calibrates 3D motion and performs 4D reconstruction of dynamic scenes using monocular video without relying on external optical flow or tracking.

Refracting Reality: Generating Images with Realistic Transparent Objects

Yue Yin (Australian National University), Dylan Campbell (Australian National University)

🎯 What it does: This paper proposes a training-free image generation method called Snellcaster, which generates images containing transparent objects by synchronizing perspective views with panoramic views under text prompts, utilizing Snell's law to achieve physically plausible refraction and reflection.

Reframing Long-Tailed Learning via Loss Landscape Geometry

Shenghan Chen (Shandong University), Xiankai Lu (Shandong University)

🎯 What it does: Propose a long-tailed learning framework based on loss landscape to address tail performance degradation while balancing overall performance of head and tail classes.

ReFTA: Breaking the Weight Reconstruction Bottleneck in Tensorized Parameter-Efficient Fine-Tuning

Jingjing Zheng (Peking University), Yankai Cao (University of British Columbia)

🎯 What it does: This paper proposes an unweighted tensor parameter-efficient fine-tuning method called ReFTA;

RefTon: Reference person shot assist virtual Try-on

Liuzhuozheng Li (University of Tokyo), Yuhui Yin (360 AI Research)

🎯 What it does: Propose a RefTon framework based on Flux, achieving virtual try-on using only the source person image and target clothing image, and enhancing texture details and realism through additional reference images (photos of the same clothing worn by others).

ReGenHOI: Unifying Reconstruction and Generation for 3D Human-Object Interaction Understanding

Miao Xu (Chinese Academy of Sciences), Zhen Lei (Chinese Academy of Sciences)

🎯 What it does: This paper proposes a unified framework that can reconstruct 3D human-object interaction states from images and generate reasonable interaction actions based on natural language.

RegFormer: Transferable Relational Grounding for Efficient Weakly-Supervised Human-Object Interaction Detection

Jihwan Park (KAIST), Hyunwoo J. Kim (KAIST)

🎯 What it does: Achieved weakly supervised human-object interaction detection (HOI) using only image-level labels for training, enabling direct inference of instance-level interaction information.

Region-Adaptive Sampling for Diffusion Transformers

Ziming Liu (National University of Singapore), Yuqing Yang (Microsoft Research)

🎯 What it does: Proposed a training-free region-adaptive sampling method called RAS, which uses Diffusion Transformers to update only the attention regions at each step while directly caching noise from other regions, significantly accelerating the text-to-image sampling process.

Region-Aware Instance Consistency Learning for Micro-Expression Recognition

Yaomin Cai (South China University of Technology Pazhou Lab), Tong Zhang (South China University of Technology Pazhou Lab)

🎯 What it does: Proposes a vertex-frame-annotation-free micro-expression recognition framework named Ra-ICL, which extracts subtle motion information using techniques such as multi-instance representation, instance region consistency, and multi-region discovery.

Region-Wise Correspondence Prediction between Manga Line Art Images

Yingxuan Li (University of Tokyo), Yusuke Matsui (University of Tokyo)

🎯 What it does: This paper proposes a Transformer-based framework for predicting region-level correspondences between raw manga line art images without any annotations or pre-segmentation.

RegionFuse: Region-Adaptive Pixel Distribution Learning for Infrared and Visible Image Fusion

Jianghan Xia (Beijing Institute of Technology), Jian Yang (Beijing Institute of Technology)

🎯 What it does: This paper proposes RegionFuse, a region-adaptive fusion network based on local pixel distribution, for infrared and visible light image fusion.

RegionRoute: Regional Style Transfer with Diffusion Model

Bowen Chen (University of Texas at Austin), Divya Kothandaraman (Dolby Laboratories)

🎯 What it does: Propose the RegionRoute framework, which can achieve precise, edge-artifact-free style transfer on individual target regions without using explicit masks.

Registration-Free Learnable Multi-View Capture of Faces in Dense Semantic Correspondence

Panagiotis P. Filntisis (Athena Research Center), Timo Bolkart (Google)

🎯 What it does: MOCHI proposes a registration-free multi-view facial capture framework that directly learns high-quality fixed-topology 3D meshes from raw multi-view scans and camera calibrations.

Regulating Rather than Constraining: Adaptive Guidance for Complex Spectral Reconstruction in Pansharpening

Zhuwei Wen (Wuhan University), Xianwei Zheng (Wuhan University)

🎯 What it does: A dual regularization framework combining hybrid data augmentation MixShuffle and hierarchical attention loss HAL is proposed to address spectral mixed regions in remote sensing image stitching, along with the design of a general dual-scale attention network DANet to enhance reconstruction accuracy and generalization.

RehearseVLA: Simulated Post-Training for VLAs with Physically-Consistent World Model

Junjin Xiao (Sun Yat-sen University), Qing Zhang (Sun Yat-sen University)

🎯 What it does: Designed and implemented RehearseVLA, a post-training framework based on a physics-consistent world model and a VLM reflector, which enables reinforcement learning through virtual simulation with minimal demonstration data, addressing issues of data scarcity, safety risks, and insufficient task completion detection.

ReHyAt: Recurrent Hybrid Attention for Video Diffusion Transformers

Mohsen Ghafoorian (Qualcomm AI Research), Amirhossein Habibian (Qualcomm AI Research)

🎯 What it does: Proposed a cyclic hybrid attention mechanism ReHyAt for video diffusion transformers, achieving a combination of local softmax and global linear attention, supporting constant memory and linear time inference.

Reinforce to Learn, Elect to Reason: A Dual Paradigm for Video Reasoning

Songyuan Yang (National University of Defense Technology), Nong Xiao (Sun Yat-sen University)

🎯 What it does: Propose a dual-paradigm RL-ER (Reinforce to Learn, Elect to Reason), which employs reinforcement learning to generate verifiable keyframe evidence and performs multi-candidate answer voting and self-inspection based on the evidence during inference, addressing the unreliability of single-channel video reasoning.

Reinforcement-Guided Synthetic Data Generation for Privacy-Sensitive Identity Recognition

Xuemei Jia (Wuhan University), Zheng Wang (Wuhan University)

🎯 What it does: Designed a synthetic data generation framework based on reinforcement learning, leveraging a pre-trained Diffusion Transformer from a general domain for domain adaptation in privacy-restricted identity recognition tasks, and enhancing the semantic consistency, diversity, and expressiveness of generated samples through multi-objective rewards.

Reinforcing Structured Chain-of-Thought for Video Understanding

Peiyao Wang (Stony Brook University), Kah Kuen Fu (Amazon)

🎯 What it does: Proposed a single-stage reinforcement learning framework SDRL, enhancing the temporal reasoning and factual consistency of video understanding models through structured chain-of-thought (Summarize → Think → Answer).

Reinforcing Video Object Segmentation to Think before it Segments

Sitong Gong (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)

🎯 What it does: Propose a video reasoning segmentation framework called Veason-R1 with a 'think first, then segment' approach, which first selects key frames, performs target localization, and completes segmentation through mask propagation.

Rejection Mixing: Fast Semantic Propagation of Mask Tokens for Efficient DLLM Inference

Yushi Ye (Shanghai Jiao Tong University), Jiangchao Yao (Shanghai Jiao Tong University)

🎯 What it does: This paper proposes a training-agnostic decoding framework called ReMix, which introduces a Continuous Mixing State between Mask and Token states and applies a Rejection Rule to iteratively refine combinatorial contradictions in parallel decoding of DLLMs, significantly improving inference speed while maintaining or even enhancing generation quality.

REL-SF4PASS: Panoramic Semantic Segmentation with REL Depth Representation and Spherical Fusion

Xuewei Li (Shanghai Dianji University), Xi Li (Shanghai Dianji University)

🎯 What it does: Propose a REL (Rectified Depth, Elevation-Gained Vertical Inclination Angle, Lateral Orientation Angle) depth representation based on cylindrical coordinates, and design Spherical-dynamic Multi-Modal Fusion (SMMF) to perform region-level dynamic fusion of RGB and REL, thereby enhancing the performance of panoramic semantic segmentation.

Rel-Zero: Harnessing Patch-Pair Invariance for Robust Zero-Watermarking Against AI Editing

Pengzhen Chen (Chinese Academy of Sciences), Weiping Wang (Chinese Academy of Sciences)

🎯 What it does: Proposes a zero-watermark framework called Rel-Zero based on the invariance of distances between patch pairs, which can generate and verify watermarks without altering the original image.

ReLaGS: Relational Language Gaussian Splatting

Yaxu Xie (German Research Center for Artificial Intelligence), Didier Stricker (German Research Center for Artificial Intelligence)

🎯 What it does: This work proposes the ReLaGS framework, achieving the unified construction of training-free multi-level language Gaussian fields and open-vocabulary 3D scene graphs, supporting tasks such as multi-level semantic queries, relationship reasoning, and relation-driven instance segmentation.

Relational Visual Similarity

Thao Nguyen (University of Wisconsin-Madison), Yuheng Li (Adobe Research)

🎯 What it does: This paper proposes and implements a new visual similarity metric called relational visual similarity, and constructs a training framework based on anonymized captions;

ReLaX: Reasoning with Latent Exploration for Large Reasoning Models

Shimin Zhang (Hong Kong Polytechnic University), Jibin Wu (Hong Kong Polytechnic University)

🎯 What it does: Propose ReLaX, an RLVR training framework that utilizes Koopman transform to analyze the hidden state dynamics of large reasoning models (LRMs), employing Dynamic Spectral Dispersion (DSD) to measure the diversity of hidden state dynamics and embedding it into the GRPO loss to regulate the exploration-exploitation balance.

Reliable Clustering Number Estimation for Contrastive Multi-View Clustering

Zhengzhong Zhu (Sichuan University), Jiangping Zhu (Sichuan University)

🎯 What it does: Propose a contrastive multi-view clustering framework RCNMC with reliable cluster number estimation, addressing the dependency of traditional methods on predefined cluster numbers and the representation degradation caused by uneven view quality.

Reliable Policy Transfer for Safety-Aware End-to-End Driving with Deep Reinforcement Learning

Uddin Md. Borhan (Shenzhen University), Jie Chen (Shenzhen University)

🎯 What it does: Propose a unified deep reinforcement learning framework that realizes safety-aware end-to-end autonomous driving through a control layer reliability interface.

Reliev3R: Relieving Feed-forward 3D Reconstruction from Multi-View Geometric Annotations

Youyu Chen (Harbin Institute of Technology Huawei), Dave Zhenyu Chen

🎯 What it does: Training a feedforward 3D reconstruction model from scratch without multi-view geometry annotations

Relightable Holoported Characters: Capturing and Relighting Dynamic Human Performance from Sparse Views

Kunwar Maheep Singh (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)

🎯 What it does: This paper proposes Relightable Holoported Characters (RHC), a light-adjustable rendering method for full-body dynamic humans based on sparse RGB views, capable of generating high-fidelity renderings under arbitrary viewpoints and environmental lighting conditions during inference using only four camera inputs captured under neutral illumination.

RelightAnyone: A Generalized Relightable 3D Gaussian Head Model

Yingyan Xu (ETH Zuerich), Studios blank

🎯 What it does: Propose the RelightAnyone method to reconstruct 3D Gaussian head models from single or multi-view images, which can be relit under arbitrary illumination.

Relightful Video Portrait Harmonization

Jun Myeong Choi (University of North Carolina at Chapel Hill), Joon-Young Lee (Adobe Research)

🎯 What it does: Propose a video harmonization framework named HarmoVid, which can seamlessly integrate foreground videos with any background videos while preserving the identity of the foreground subject, achieving high-quality video synthesis with consistent lighting, shadows, tone, and background matching.

ReManNet: A Riemannian Manifold Network for Monocular 3D Lane Detection

Chengzhi Hong (Wuhan University), Bijun Li (Wuhan University)

🎯 What it does: Propose a 3D lane detection framework called ReManNet based on the Riemannian plane assumption of the road surface, which can integrate visual features with Gaussian descriptors on SPD (symmetric positive definite) matrices and perform information fusion through a gating mechanism.

ReMatch: Boosting Representation through Matching for Multimodal Retrieval

Qianying Liu (University of Glasgow), Paul Henderson (Xiaohongshu Inc.)

🎯 What it does: ReMatch achieves end-to-end representation learning for image-text retrieval tasks by introducing a chat-style generative matching phase and multi-learnable tokens for multi-vector fusion on multimodal large language models (MLLM).

RemedyGS: Defend 3D Gaussian Splatting Against Computation Cost Attacks

Remedying Target-Domain Astigmatism for Cross-Domain Few-Shot Object Detection

Yongwei Jiang (Huazhong University of Science and Technology), Ruixuan Li (Huazhong University of Science and Technology)

🎯 What it does: In cross-domain few-shot object detection, the authors propose a central-peripheral attention refinement framework to address the target domain astigmatism problem;

ReMoE: Region-Mixture Experts for Adversarially-Robust Vision Transformers

Qinghao Zhong (Beijing Institute of Technology), Guangming Lu (Shenzhen University)

🎯 What it does: Proposed a Region-aware Mixture-of-Experts (ReMoE) module, replacing the FFN in Vision Transformers with multi-granularity experts (global, center, local), and achieving regional consistency through attention-guided P2R and R2P routing to enhance adversarial robustness.

ReMoGen: Real-time Human Interaction-to-Reaction Generation via Modular Learning from Diverse Data

Yaoqin Ye (ShanghaiTech University), Yuexin Ma (Chinese University of Hong Kong)

🎯 What it does: Proposes ReMoGen, a modular framework for real-time human interactive-to-reactive generation, capable of generating high-quality, coherent future motions of characters in both single-domain and mixed-modal interaction scenarios.

ReMoRa: Multimodal Large Language Model based on Refined Motion Representation for Long-Video Understanding

Daichi Yashima (Keio University), Komei Sugiura (Keio University)

🎯 What it does: Develop a long video multi-modal large language model called ReMoRa, which works directly on compressed video streams, leveraging the appearance information from I-frames and compressed domain motion vectors. The block-level motion vectors are denoised and refined through a Refined Motion Representation (RMR) module, followed by a Hierarchical Motion State Space (HMSS) module that performs linear time-series modeling on GOP-level motion sequences. Finally, the fused features are input into a pre-trained LLM for long video understanding and reasoning.

ReMoT: Reinforcement Learning with Motion Contrast Triplets

Cong Wan (Xi'an Jiaotong University), Yihong Gong (Xi'an Jiaotong University)

🎯 What it does: Propose ReMoT, a unified training paradigm that enhances the performance of vision-language models (VLM) in spatiotemporal consistency reasoning by constructing a motion contrast triplet dataset called ReMoT-16K and combining it with Group Relative Policy Optimization (GRPO).

Remote Sensing Image Super-Resolution for Imbalanced Textures: A Texture-Aware Diffusion Framework

Enzhuo Zhang (Nanjing University), Pengfeng Xiao (Nanjing University)

🎯 What it does: This paper proposes the TexADiff framework, addressing the texture imbalance problem in remote sensing image super-resolution by using the texture relative density map (RTDM) for spatial conditioning, loss weighting, and sampling scheduling.

Render-to-Adapt: Unsupervised Personal Adaptation for Gaze Estimation

Yangshi Ge (Beihang University), Feng Lu (Beihang University)

🎯 What it does: Proposes an unsupervised personalized gaze estimation method, Render-to-Adapt (R2A), which self-supervisedly calibrates pre-trained models through a differentiable renderer and Render-Cycle Consistency.

RenderFlow: Single-Step Neural Rendering via Flow Matching

Shenghao Zhang (Disney Research Studios), Yang Zhang (Disney Research Studios)

🎯 What it does: Propose RenderFlow, a single-step neural rendering framework based on flow matching, which can directly generate high-quality illumination images from G-buffer and support sparse keyframe guidance and inverse rendering.

Reparameterized Tensor Ring Functional Decomposition for Multi-Dimensional Data Recovery

Yangyang Xu (Hunan Normal University), Chao Wang (Southern University of Science and Technology)

🎯 What it does: Proposed Tensor Ring Functional Decomposition (TRFD) and achieved continuous-domain high-order data recovery through reparameterization (RepTRFD), addressing the grid limitations and spectral bias issues of traditional Tensor Ring (TR).

Representation-Steered Incremental Adapter-Tuning for Class-Incremental Learning with Pre-Trained Models

Jiarui Zhao (Chinese Academy of Sciences), Yongjun Xu (Chinese Academy of Sciences)

🎯 What it does: Proposes Representation-Steered Incremental Adapter Tuning (RSIAT), a class-incremental learning framework based on pre-trained models, which employs shared adapters and achieves stable and efficient incremental learning through representation-guided loss, residual autoencoder projection, and orthogonal loss.

Representing 3D Faces with Learnable B-Spline Volumes

Prashanth Chandran (Google), Timo Bolkart (Google)

🎯 What it does: Propose a novel 3D facial representation called CUBE, which generates high-fidelity facial meshes using learnable high-dimensional B-Spline control features and a lightweight MLP.

Repurposing 3D Generative Model for Autoregressive Layout Generation

Haoran Feng (Beihang University), Lu Sheng (Beihang University)

🎯 What it does: Propose the LaviGen framework, utilizing 3D generative models to achieve autoregressive native 3D layout generation;

ResAD: Normalized Residual Trajectory Modeling for End-to-End Autonomous Driving

Zhiyu Zheng (Wuhan University), Lefei Zhang (Wuhan University)

🎯 What it does: Propose an end-to-end autonomous driving framework ResAD, which uses inertial reference trajectories and residual modeling, combining point-level residual normalization and inertial reference perturbation to achieve multi-modal planning.

ReSAM: Refine, Requery, and Reinforce: Self-Prompting Point-Supervised Segmentation for Remote Sensing Images

Muhammad Naseer Subhani (Independent Researcher)

🎯 What it does: Proposed a point-supervised self-prompting framework called ReSAM, which gradually converts sparse point annotations into high-quality pseudo masks through a Refine-Requery-Reinforce cycle, achieving unsupervised domain adaptation for remote sensing images.

ResCa: Residual Caching for Diffusion Transformers Acceleration

Haipeng Fang (Institute of Computing Technology, Chinese Academy of Sciences), Sheng Tang (Institute of Computing Technology, Chinese Academy of Sciences)

🎯 What it does: Propose a training-free residual caching framework called ResCa, which utilizes the true denoising residuals of selected proxy tokens in each cluster to guide the simulated denoising of the remaining tokens, thereby accelerating the inference of diffusion transformers.

ReScene4D: Temporally Consistent Semantic Instance Segmentation of Evolving Indoor 3D Scenes

Emily Steiner (Stanford University), Iro Armeni (Stanford University)

🎯 What it does: A 4D semantic instance segmentation method that simultaneously performs semantic instance segmentation, instance recognition, and cross-temporal association is proposed for 3D scans in indoor environments with sparse temporal changes.

ResDiT: Evoking the Intrinsic Resolution Scalability in Diffusion Transformers

Yiyang Ma (Beijing University of Posts and Telecommunications), Jianqin Yin (Beijing University of Posts and Telecommunications)

🎯 What it does: Propose ResDiT, a no-training, high-resolution image synthesis framework based on pre-trained Diffusion Transformer.

Residual Connections Harm Generative Representation Learning

Xiao Zhang (University of Chicago), Michael Maire (University of Chicago)

🎯 What it does: Introduce layer-wise decayed identity shortcuts into self-supervised generative models (Masked Autoencoders and Diffusion Models) to reduce the direct transmission of low-level features in residual networks, thereby enhancing semantic feature learning and generation quality.

Residual Decoder Adapter: ID-Preserving Tokenizer Adaption for Autoregressive Text Rendering

Dongxing Mao (Central South University), Jingru Tan (Microsoft Research)

🎯 What it does: Propose a Residual Decoder Adapter (RDA), which improves text rendering performance without modifying existing reversible visual tokenizers by adding a shared ID codebook and residual decoder to them.

Residual Decoding: Mitigating Hallucinations in Large Vision-Language Models via History-Aware Residual Guidance

Xinrong Chen (Beijing University of Posts and Telecommunications), Ngai Wong (University of Hong Kong)

🎯 What it does: Proposes a no-training-cost, plug-and-play residual decoding (ResDec) method to suppress hallucinations caused by language bias during the generation process of large vision-language models (LVLMs);

Residual Diffusion Bridge Model for Image Restoration

Hebaixu Wang (Wuhan University), Bo Du (Wuhan University)

🎯 What it does: Propose the Residual Diffusion Bridge Model (RDBM) for unified handling of various image restoration tasks.

Residual Primitive Fitting of 3D Shapes with SuperFrusta

Aditya Ganeshan (Brown University), Daniel Ritchie (Brown University)

🎯 What it does: The paper proposes a method that converts 3D shapes into editable, compact primitive assemblies using the SuperFrustum primitive and the ResFit algorithm.

ResiHMR: Residual-Limb Aware Single-Image 3D Human Mesh Recovery for Individuals with Limb Loss

Jiaying Ying (University of Queensland), Xin Yu (University of Adelaide)

🎯 What it does: Proposes ResiHMR, a 3D human mesh reconstruction framework for amputee individuals from a single image, capable of explicitly recovering residual limb surfaces and adaptively adjusting topology.

Resolving Endpoint Underfitting in Diffusion Bridges via Noise Alignment

Yurong Gao (University of Chinese Academy of Sciences), Xinmin Qiu (University of Chinese Academy of Sciences)

🎯 What it does: This paper proposes the Noise-Aligned Diffusion Bridge (NADB), addressing the underfitting issue at the target endpoint of existing diffusion bridges by redesigning the noise-aligned bridge model.

Resolving Evidence Sparsity: Agentic Context Engineering for Long-Document Understanding

Keliang Liu (Fudan University), Lihua Zhang (Fudan University)

🎯 What it does: Proposes SLEUTH, a no-training, plug-and-play multi-agent framework that constructs highly dense, cross-modal evidence context through a coarse-to-fine process, utilizing modules such as retrieval, clue discovery, page screening, difficulty assessment, and core decision-making, to achieve long document question answering.

Resolving the Identity Crisis in Text-to-Image Generation

Shubhankar Borse (Qualcomm AI Research), Fatih Porikli (Qualcomm AI Research)

🎯 What it does: To address the multi-human identity confusion problem in text-to-image generation, the DISCO framework is proposed, which refines multi-human generation models using reinforcement learning, ensuring diversity and accurate quantity of identities in the generated images.

Resolving the Stability-Plasticity Dilemma in Reinforcement Learning via Complementary Continual Critics

Bo Sun (Sun Yat-sen University), Luntong Li (Peng Cheng Laboratory)

🎯 What it does: Proposed a Continual Dual-Critic with Cross-Attention (CD-CCA) framework that leverages dual Critics combined with Continual Backpropagation (CBP) and Elastic Weight Consolidation (EWC), and dynamically fuses value estimates through Cross-Attention to address the plasticity-stability dilemma in visual reinforcement learning.

Restore Text First, Enhance Image Later: Two-Stage Scene Text Image Super-Resolution with Glyph Structure Guidance

Minxing Luo (Nankai University), Jian Yang (Nankai University)

🎯 What it does: Propose the TiGeSR two-stage framework, first using a diffusion model to accurately restore the stroke structure of scene text, then using the restored text structure as a condition to guide full-image super-resolution, balancing text readability and overall visual quality.

Restore, Assess, Repeat: A Unified Framework for Iterative Image Restoration

I-Hsiang Chen (Samsung AI Center), Brais Martinez (Samsung AI Center)

🎯 What it does: Proposes a unified Restore-Assess-Repeat (RAR) framework that iteratively restores unknown or compound degraded images by jointly utilizing image quality assessment (IQA) in the latent space and generative image restoration (IR), with the entire process trainable end-to-end and adaptively terminated during inference.

RetFormer: Multimodal Retrieval for Enhancing Image Recognition

Tianrui Yu (Zhejiang University), Hongzhi Wang (Zhejiang University)

🎯 What it does: Propose a multi-modal retrieval-enhanced visual recognition framework called RetFormer, which retrieves similar samples from an external image-text knowledge base and fuses image and text features to improve robustness in long-tail recognition and noisy label learning.