paperlist.ai

AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

ECCV 2026

TryOnCrafter: Unleashing Camera Trajectories for Realistic Video Virtual Try-on via a Renderable 4D Try-on Proxy

Hao Sun, Hao Yan, Mengting Chen, Quanjian Song, Yu Li, Juan Cao, Jinsong Lan, Xiaoyong Zhu, Bo Zheng, Sheng Tang

While Video Virtual Try-on (VVT) has achieved remark-able progress in synthesizing realistic garment overlays on dynamic sub-jects, existing paradigms remains fundamentally constrained by a pas-sive dependency on source camera trajectories, failing to accommodatethe requisite interactive freedom for omnidirectional viewpoint explo-ration. To address this limitation, we define a pioneering research fron-tier: Camera-controllable Video Virtual Try-on (CaM-VVT). Unlike con-ventional VVT, CaM-VVT not only necessitates viewpoint-agnostic tex-ture hallucination but also strict structural synchronization between non-rigid human dynamics and background contexts under arbitrary, un-constrained camera movements. To tackle these challenges, we presentTryOnCrafter, the first unified DiT-based framework specifically archi-tected for the CaM-VVT task. Departing from implicit pixel-space ma-nipulation, we introduce a Renderable 4D Try-on Proxy that explicitlydecouples the human subject from the environment. This is achieved bydistilling high-fidelity 2D try-on priors into a clothed 3DGS-based avatar,which is subsequently animated via SMPL-X sequences and metric-alignedinto a reconstructed background point cloud. This proxy establishes a ro-bust structural foundation with superior texture density and motion in-tegrity. Our Proxy-Anchored Video DiT leverages this robust structuralfoundation as a primary geometric anchor, ensuring that the synthesizedphotorealistic videos are strictly constrained by prescribed trajectoriesand physically plausible deformations. Benefiting from the inherent ed-itability of the 4D proxy, TryOnCrafter facilitates diverse downstreamapplications, including human relocalization, “bullet time” effects, and360-degree orbital viewing. Extensive experiments on our establishedCaM-VVTBench demonstrate that TryOnCrafter significantly outper-forms existing baselines in preserving structural consistency and garmentidentity across complex camera maneuvers.

ECCV 2026

PhysConvex: Physics-Informed Dynamic Convex Fields for Reconstruction and Simulation

Dan Wang, Xinrui Cui, Serge Belongie, Ravi Ramamoorthi

Reconstructing deformable objects from video requires a 4Drepresentation that should preserve geometry, explain motion throughphysics, and generalize to future or new physical conditions. Existingdynamic NeRF and Gaussian methods achieve strong view synthesis,yet their voxel or ellipsoidal primitives are primarily designed for ap-pearance rendering and typically driven by centers or predefined par-ticle bindings, limiting physically meaningful non-uniform deformationand sharp boundary evolution. We present PhysConvex, a physics-informed dynamic convex field for video-based reconstruction, physicalsystem identification, and simulation. PhysConvex represents a dynamicobject as material-space deformable convex primitives whose boundaryis advected by physical dynamics. The boundary-driven convex expressesnon-uniform deformation and evolving active supports while serving si-multaneously as a rendering element, deformation carrier, and physicalsupport where mass, elastic response, forces, and contacts are evalu-ated. We further introduce a mesh-free reduced-order convex simulatorin which neural skinning modes define physics-based deformation basesdirectly over deformable convex supports. Experiments show improveddynamic and physical reconstruction, efficiency, future prediction, andgeneralization to changed materials, forces, and boundary conditions.

ECCV 2026

From Sparse to Dense: Multi-View GRPO for Flow Models via Augmented Condition Space

Jiazi Bu, Pengyang Ling, Yujie Zhou, Yibin Wang, Yuhang Zang, Tianyi Wei, Xiaohang Zhan, Jiaqi Wang, Tong Wu, Xingang Pan, Dahua Lin

Group Relative Policy Optimization (GRPO) has emergedas a powerful framework for preference alignment in text-to-image (T2I)flow models. However, we have observed that the standard paradigmthat evaluates a group of generated samples against a single conditionsuffers from insufficient exploration of inter-sample relationships, con-straining both alignment efficacy and performance ceilings. To addressthis sparse single-view evaluation scheme, we propose Multi-View GRPO(MV-GRPO), a novel algorithm that enhances relationship explorationby augmenting the condition space to create a dense multi-view rewardmapping. Specifically, for a group of samples generated from one prompt,MV-GRPO leverages a flexible Condition Enhancer to generate seman-tically adjacent yet diverse captions. These captions enable multi-viewadvantage re-estimation, capturing diverse semantic attributes and pro-viding richer optimization signals. By deriving the probability distribu-tion of the original samples conditioned on these new captions, they canbe incorporated into the training process without costly sample regen-eration. Extensive experiments demonstrate that MV-GRPO achievessuperior alignment performance over state-of-the-art methods.

ECCV 2026

Deep Noise Label Learning via Effective Rank Reduction

Changyi Ma, Zihan Fang, Lihua Zhou, Tao Li, Wenyu Liu, Runsheng Yu, Xuan Song

Learning with noisy labels in large-scale classification re-mains challenging. Existing forward-correction methods estimate class-dependent label noise via a transition matrix, but this matrix is typ-ically assumed to be full-rank and unconstrained. However, real-worldlabel noise is typically structured and exhibits low-rank properties dueto semantic clustering, especially when the number of classes is large.This mismatch between modeling assumptions and the intrinsic struc-ture of label noise leads to unstable optimization and degraded general-ization performance. Motivated by this observation, we propose a novelLow-Effective-Rank Noisy-Label Learning (LENL) framework, whichexplicitly constrains the transition matrix via nuclear-norm regulariza-tion. This low-rank inductive bias captures structured confusion pat-terns while suppressing spurious noise modes, yielding both statisticalefficiency and improved generalization. We further provide theoreticalanalysis establishing the benefits of rank-constrained estimation understructured noise assumptions. To enable stable and scalable optimiza-tion, we introduce a Newton–Schulz optimizer tailored for label-noiselearning, which approximates the nuclear norm through decomposition-free iterative updates, avoiding the numerical instability and computa-tional overhead of explicit singular value decomposition. Extensive exper-iments on CIFAR-100, CIFAR-100N, ImageNet-1K, and Noisy Ostracodsdemonstrate consistent and significant improvements over state-of-the-art label-noise learning methods.

ECCV 2026

SPEAR: A Simulator for Photorealistic Embodied AI Research

Mike Roberts, Renhan Wang, Rushikesh Zawar, Rachith Dey-Prakash, Quentin Leboutet, Stephan Richter, Matthias Muller, German Ros, Rui Tang, Stefan Leutenegger, Yannick Hold-Geoffroy, Kalyan Sunkavalli, Vladlen Koltun

Interactive simulators have become powerful tools for train-ing embodied agents and generating synthetic visual data, but existingphotorealistic simulators suffer from limited generality, programmabil-ity, and rendering speed. We address these limitations by introducingSPEAR: A Simulator for Photorealistic Embodied AI Research. At itscore, SPEAR is a Python library that can connect to, and program-matically control, any Unreal Engine (UE) application via a modularplugin architecture. SPEAR exposes over 14K unique UE functions toPython, representing an order-of-magnitude increase in programmablefunctionality over existing UE-based simulators. Additionally, a singleSPEAR instance can render 1920×1080 photorealistic beauty images di-rectly into a user’s NumPy array at 73 frames per second – an order ofmagnitude faster than existing UE plugins – while also providing groundtruth image modalities that are not available in any existing UE-basedsimulator (e.g., a non-diffuse intrinsic image decomposition, material IDs,and physically based shading parameters). Finally, SPEAR introduces anexpressive high-level programming model that enables users to specifycomplex graphs of UE work with arbitrary data dependencies amongwork items, and to execute these graphs deterministically within a sin-gle UE frame. We demonstrate the utility of SPEAR through a diversecollection of example applications: controlling multiple embodied agentswith distinct action spaces (e.g., humans, cars, and robots) across sev-eral in-the-wild UE projects; rendering photorealistic city-scale environ-ments; manipulating UE’s procedural content generation systems; ren-dering synchronized multi-view images of detailed human faces; coordi-nating an interactive co-simulation with the MuJoCo physics simulator;and editing scenes with natural language via an AI coding assistant.

ECCV 2026

ActionParty: Multi-Subject Action Binding in Generative Video Games

Alexander Pondaven, Ziyi Wu, Igor Gilitschenski, Philip Torr, Sergey Tulyakov, Fabio Pizzati, Aliaksandr Siarohin

Recent advances in video diffusion have enabled the development of “world models” capable of simulating interactive environments. However, these models are largely restricted to single-agent settings, failing to control multiple agents simultaneously in a scene. In this work, we tackle a fundamental issue of action binding in existing video diffusion models, which struggle to associate specific actions with their corresponding subjects. For this purpose, we propose ActionParty, an action controllable multi-subject world model for generative video games. It introduces subject state tokens, i.e. latent variables that persistently capture the state of each subject in the scene. By jointly modeling state tokens and video latents with a spatial biasing mechanism, we disentangle global video frame rendering from individual action-controlled subject updates. We evaluate ActionParty on the Melting Pot benchmark, demonstrating the first video world model capable of controlling up to seven players simultaneously across 46 diverse 2D game environments. Our results show significant improvements in action-following accuracy and identity consistency, while enabling robust autoregressive tracking of subjects through complex interactions.

ECCV 2026

DP-BOA: Dirichlet-Process Birth-or-Assign for On-the-Fly Category Discovery

Peiyan Gu, Zixin Teng, Xuming He

On-the-fly category discovery requires deciding for each in-coming test sample whether to assign it to an existing category or spawna new one. Existing methods typically implement this decision throughmatching-based heuristics, such as radius- or hash-based rules. While ef-fective in practice, these methods usually treat category birth implicitlyas a fallback when no existing category matches confidently, rather thanas an explicit alternative supported by its own statistical evidence. Toaddress this, we propose DP-BOA, a posterior-predictive decision frame-work based on an online Dirichlet-process Gaussian mixture model witha Normal–Inverse–Wishart prior. During training, we use labeled datato calibrate a shared NIW prior over category Gaussians and warm-startthe known-category posteriors. At test time, for each incoming sample,DP-BOA compares the posterior predictive evidence for assignment toexisting categories against the evidence for spawning a new category in-duced by the DP prior, and then updates category statistics online afterthe decision. The method captures anisotropic category geometry andnaturally adapts decision confidence as evidence accumulates. Acrossstandard OCD benchmarks, DP-BOA consistently outperforms strongbaselines and delivers particularly strong novel-class discovery perfor-mance while maintaining competitive known-class accuracy. The projectpage is available at DP-BOA.

ECCV 2026

PADFormer: Pose-agnostic Anomaly Detection from Sparse View Images

Ruiqi Wang, Yiming Qian, Fenggen Yu, Yuxuan Lu, Dakuo Wang, Hao Richard Zhang, Jing Huang

Pose-agnostic Anomaly Detection (PAD) remains challeng-ing as anomalies can appear under arbitrary viewpoints, requiring meth-ods to handle significant pose variations. Existing approaches rely oncomplex 3D reconstruction, which are computationally expensive and re-quire extensive multi-view data. We propose PADFormer, a novel image-space approach that leverages Vision Transformer (ViT) to directly re-construct anomaly-free versions of query images while preserving poseinformation. Our key insight is to adapt cross-view masked reconstruc-tion for anomaly detection through training exclusively on normal data,combined with dynamic patch selection and spatial alignment mecha-nisms that enable effective learning from sparse reference views undersignificant pose variations. During inference, we perform multiple for-ward passes with different masking patterns to generate an ensembleof anomaly-free reconstructions, ensuring comprehensive coverage of thequery image. Anomalies are detected by comparing these reconstructionswith the query image. PADFormer achieves state-of-the-art results on thePAD benchmark while maintaining comparable performance on classicfew-shot anomaly detection (FSAD) tasks, demonstrating superior effi-ciency and generalization without requiring 3D reconstruction.

ECCV 2026

Push–Pull Attentional Anchoring for Diffusion Concept Erasure

Nattanat Chatthee, Tagon Sompong, Ekapol Chuangsuwanich, Supasorn Suwajanakorn

Rapid advances in diffusion models have raised concernsabout privacy, harmful content, and copyright violations. Concept era-sure methods aim to address these risks by removing undesirable con-cepts from pretrained models without full retraining. However, existingapproaches often degrade non-target generation quality through heuristicstrength scaling (e.g., classifier-free guidance or negative prompt scaling),leading to uncontrolled and excessive semantic drift. To address this, wepropose Push–Pull Attentional Anchoring (PPAA), a mechanism in thecross-attention space that displaces target representations while anchor-ing the scale of erasure via bounded cosine-similarity constraints. Byusing a relative similarity ratio instead of an absolute difference, ouradaptive threshold represents a meaningful percentage of semantic re-tention, facilitating a single value to be applied uniformly across layers.We conduct extensive experiments across multiple concept categories, in-cluding celebrities, artistic styles, nudity, and intellectual property. Ourmethod achieves favorable trade-offs against prior approaches on StableDiffusion v1.4 and demonstrates its applicability to modern architecturessuch as FLUX.

ECCV 2026

Stream-DiffVSR: Low-Latency Streamable Video Super-Resolution via Auto-Regressive Diffusion

Hau-Shiang Shiu, Chin-Yang Lin, Zhixiang Wang, Chi-Wei Hsiao, Po-Fan Yu, Yu-Chih Chen, Yu-Lun Liu

Diffusion-based video super-resolution (VSR) methods deliver strong perceptual quality but are often unsuitable for latency-sensitive scenarios due to reliance on future frames and expensive multi-step denoising. We propose Stream-DiffVSR, a causally conditioned diffusion framework for efficient online VSR. Operating strictly on past frames, Stream-DiffVSR integrates a four-step distilled denoiser for fast inference, an Auto-regressive Temporal Guidance (ARTG) module that injects motion-aligned cues during latent denoising, and a lightweight temporalaware decoder with a Temporal Processor Module (TPM) to enhance detail and temporal coherence. Unlike chunk-wise streaming inference, our strictly frame-by-frame causal design avoids sequence-level waiting, substantially reducing time-to-first-frame and end-to-end latency. Stream-DiffVSR processes 720p frames in 0.328 seconds on an RTX 4090 and consistently outperforms prior diffusion-based baselines. Compared with the online state-of-the-art TMP, it improves perceptual quality (LPIPS +0.095). Compared with prior diffusion-based VSR methods such as MGLD-VSR [104], it reduces per-frame runtime by over 130×. Moreover, Stream-DiffVSR substantially lowers time-to-firstframe for diffusion-based VSR, reducing initial delay from over 4600 seconds to 0.328 seconds, making diffusion-based VSR markedly more practical for low-latency online and streaming deployment. Project page: https://jamichss.github.io/stream-diffvsr-project-page/

ECCV 2026

RADmesh: Remesh-Aware Mesh Deformation

Nam Anh Dinh, Itai Lang, Oded Stein, Rana Hanocka

We propose a remeshing-enhanced method for generativelydeforming shapes with visual losses. It is intuitive that sufficiently dras-tic deformations of a mesh without changing its triangulation can easilycompromise element quality, even if such large geometry changes maybe semantically desired. Shape deformation methods could thus bene-fit from changing the triangulation; however, this is not done by mostgenerative, text-based, visually-supervised mesh deformation methods.Remeshing is a discrete operation, proven to be especially challengingto couple with the notoriously noisy supervision signal provided by vi-sual losses. We propose a vertex-based deformation optimization quantitycapable of large deformations and robustness to such noise; we period-ically remesh using an isotropic remesher that interpolates and carriesforward the deformation optimization state. This enables continuous,geometry-informed progress in coarse-to-fine addition of resolution. Theresulting shapes’ triangulations fit their optimized geometry and haveneat isotropic elements. Further, our method is localizable, able to grownew features on a base shape with expressive detail, leaving the rest un-changed. We showcase the effectiveness of our method on a variety ofshapes and prompts, both local and global deformations, and demon-strate its superior visual quality and triangle efficiency. Our project pageis at https://threedle.github.io/radmesh.

ECCV 2026

Generalizable Neural Reconstruction of High-Fidelity Surfaces via Sparse Volumetric Representations

Aoxiang Fan, Corentin Dumery, Nicolas Talabot, Ming Xu, Hieu Le, Pascal Fua

Neural implicit representations have recently achieved im-pressive results in novel view synthesis and multi-view 3D reconstruc-tion, yet both NeRF- and Gaussian Splatting-based methods require per-scene optimization, which makes them inefficient. Generalizable NeuralSurface Reconstruction (GNSR) methods have been proposed to removethis need by learning feature representations directly predicted from in-put images. However, their typical reliance on dense feature volumesseverely limits achievable resolution and fidelity due to prohibitive mem-ory costs. We introduce Sparse Volumetric Reconstruction (SVRecon),a new GNSR framework that unlocks high-resolution, memory-efficientreconstruction through learned occupancy-driven sparsity, in a more ef-fective way than earlier approaches to introducing sparsity in GNSRs.Our approach uses a nested two-stage architecture: (1) an occupancyprediction network that identifies surface-containing voxels, and (2) ahigh-resolution sparse volume rendering framework defined only withinthese occupied regions, together with specialized sparsified algorithmsfor ray sampling, feature aggregation, and querying. This design enablesfine-grained surface reconstruction while avoiding the heavy memoryfootprint of dense grids. SVRecon operates at resolutions up to 5123on standard 32GB hardware—substantially higher than prior generaliz-able methods—and delivers smoother and more precise reconstructionsacross diverse datasets, particularly in sparse-view settings.

ECCV 2026

Infinite-Homography as Robust Conditioning for Camera-Controlled Video Generation

Min Jung Kim, Jeongho Kim, Hoiyeong Jin, Junha Hyung, Choo Jaegul

Recent progress in video diffusion models has spurred grow-ing interest in camera-controlled novel-view video generation for dy-namic scenes, aiming to provide creators with cinematic camera con-trol capabilities in post-production. A key challenge in camera-controlledvideo generation is ensuring fidelity to the specified camera pose, whilemaintaining view consistency and reasoning about occluded geometryfrom limited observations. To address this, existing methods either traintrajectory-conditioned video generation model on trajectory–video pairdataset, or estimate depth from the input video to reproject it alonga target trajectory and generate the unprojected regions. Nevertheless,existing methods struggle to generate camera-pose–faithful, high-qualityvideos for two main reasons: (1) reprojection-based approaches are highlysusceptible to errors caused by inaccurate depth estimation; and (2)the limited diversity of camera trajectories in existing datasets restrictslearned models. To address these limitations, we present InfCam, a depth-free, camera-controlled video-to-video generation framework with highpose fidelity. The framework integrates two key components: (1) infi-nite homography warping, which encodes 3D camera rotations directlywithin the 2D latent space of a video diffusion model. Conditioningon this noise-free rotational information, the residual parallax term ispredicted through end-to-end training to achieve high camera-pose fi-delity; and (2) a data augmentation pipeline that transforms existing syn-thetic multiview datasets into sequences with diverse trajectories and fo-cal lengths. Experimental results demonstrate that InfCam outperformsbaseline methods in camera-pose accuracy and visual fidelity, generaliz-ing well from synthetic to real-world data.

ECCV 2026

PuzLM: Solving Jigsaw Puzzles with Sequence-to-Sequence Language Models

Gur Elkin, Ofir I Shahar, Ohad Ben-Shahar

Square jigsaw puzzles are typically solved by visually match-ing piece images to recover the original layout. This work introducesPuzLM, an alternative perspective that recasts jigsaw reassembly asa discrete sequence-to-sequence (Seq2Seq) problem, inspired by naturallanguage representations. We design an efficient puzzle quantization pro-cedure that transforms each piece into a short sequence of discrete tokens,enabling the direct application of standard Seq2Seq language models aspowerful jigsaw solvers. Our approach demonstrates that accurate puzzlereconstruction can be achieved through purely symbolic reasoning overdiscrete representations, improving state-of-the-art performance even onpuzzles with eroded boundaries or missing pieces.

ECCV 2026

Diffusion Integrated Gradients: Controllable Path Generation for Flexible Feature Attribution

Soyeon Kim, Kyowoon Lee, Jaesik Choi

Path-based attribution methods such as Integrated Gradi-ents (IG) are widely adopted for their strong axiomatic properties andeffectiveness in attributing model predictions to input features by inte-grating gradients along a path from a baseline to the input. However,the choice of the attribution path largely affects the quality of explana-tions, and existing approaches rely on fixed or hand-crafted paths thatoften produce noisy or distorted attributions. To address this limitation,we propose Diffusion Integrated Gradients (DiffIG), a novel method thatreformulates path generation as a conditional generative modeling prob-lem. DiffIG first trains a diffusion model to learn a distribution overpaths generated from a Stick-Breaking Process, then employs guidedsampling to embed user guidance during the sampling procedure. Wedemonstrate that DiffIG quantitatively matches or outperforms existingpath-based methods, achieving perceptually aligned explanations. Thiswork introduces a new generative perspective for flexible, inference-timecontrollable Explainable Artificial Intelligence (XAI) methods.

ECCV 2026

Accurate Zero-shot Quantization via Hierarchical Teacher-Assistant Distillation

Wonjin Cho, Jeongin Yun, U Kang

How can we bridge the numerical chasm between full-precisionCNN models and highly quantized targets when original training dataare unavailable? As the demand for deploying deep neural networkson resource-constrained edge hardware grows, achieving extreme modelcompression without compromising the privacy of training data has be-come a critical challenge. However, direct transitions to ultra-low-bitregimes introduce extreme optimization instability and activation col-lapse, where high quantization noise causes deep layers to lose their dis-criminative capacity and representational power.In this paper, we propose Zest (ZSQ Enhancement via Stepping-stoneTransition), an accurate zero-shot quantization framework for CNNs thatbridges the precision gap between full-precision models and ultra-low-bit targets through a unified two-stage distillation and reconstructionpipeline. Rather than attempting a direct and unstable jump from fullprecision to an ultra-low bit quantized model, Zest introduces a stableintermediate bit stage to tame the numerical distribution and regular-ize the model. Furthermore, we propose a two-phase distillation strategythat initializes high-fidelity semantic skeletons from the full-precisionmodel and subsequently recalibrates them through an intermediate bitstage to ensure calibration data are numerically aligned with quantized-domain constraints. Experimental results demonstrate that this progres-sive transition leads to better distillation, achieving 70.50% top-1 accu-racy on ResNet-18 at W 4A4 precision and showing higher performancecompared to existing SOTA methods in various models.

ECCV 2026

FontCopilot: Towards Generalist Multimodal Large Language Models for Holistic Chinese Font Engineering

Yu Liu, Yang Liu, Ying Gao, Yang Ding, Cunrui Wang*

Chinese font production is a complex task that balances vi-sual consistency, vector editability, and topological integrity at a massivecharacter scale. Existing fragmented toolchains, relying on manual rules,often lead to geometric redundancy and defects. This paper introducesFontCopilot, a multimodal large language model that performs raster-to-vector conversion, Bézier curve sparsi_x001C_cation, defect marking, and geo-metric correction within a single model. We elevate font engineering fromexplicit geometric rule constraints to structured glyph code understand-ing and rewriting, serializing Bézier curves into SVG XML. This enablesthe model to perform topological reorganization in the code space andinternalize implicit designer priors as structured reasoning constraints.FontCopilot integrates few-shot reference guidance and zero-shot gen-eralization mechanisms, balancing speci_x001C_c style transfer and universalstandard reconstruction. To advance the development of a uni_x001C_ed fontengineering model, we constructed the CFCopilot-8M dataset, covering300 font types and 8 million glyph samples. Under progressive multi-taskhybrid training, the glyphs processed by FontCopilot show consistentimprovements in editability, contour compactness, and defect correction,validating its industrial-grade font production value.

ECCV 2026

Visible Yet Unrecognizable: Frequency-Selective Facial Privacy via Attention

Atul Kumar, Akshay Agarwal

Facial privacy has become a critical concern as unauthorized large-scale image scraping enables malicious actors to collect personal face images without consent and exploit them for training unauthorized recognition systems. Protecting facial privacy requires modifying a face image such that automated recognition systems fail to identify the individual, while the visual appearance remains unchanged to human observers, ensuring the image retains its social utility. However, existing methods fail to satisfy both requirements; they either irreversibly degrade visual quality, rendering images impractical, or alter facial identity to such an extent that the protected image bears no resemblance to the original subject. To address this fundamental privacy-utility tradeoff, we propose MIRAGE (Multi-scale Identity Removal via Attention-Guided Encoding), a frequency-selective facial privacy framework grounded in the observation that identity-discriminative features reside predominantly in low-frequency image components, while visual appearance is encoded in high-frequency components. We further introduce FAC (FrequencyAware Consistency) Loss, which jointly enforces identity separation and visual fidelity during training. A comprehensive evaluation across three benchmark datasets and nine state-of-the-art (SOTA) deep face recognition (DFR) models demonstrates that MIRAGE achieves robust facial privacy protection while preserving visual appearance, successfully bridging the privacy-utility gap left unresolved by existing methods. The code is available at: https://github.com/atulkr05/MIRAGE.git.

ECCV 2026

Visual Prompt Discovery via Semantic Exploration

Jaechang Kim, Yotaro Shimose, Zhao Wang, Kuang-Da Wang, Jungseul Ok, Shingo Takamatsu

Large Vision-Language Models (LVLMs) encounter significant challenges in image understanding and visual reasoning, leading to critical perception failures. Visual prompts, which incorporate image manipulation code, have shown promising potential in mitigating these issues. While visual prompts have emerged as a promising direction, previous methods for visual prompt generation have focused on tool selection rather than diagnosing and mitigating the root causes of LVLM perception failures. Because of the opacity and unpredictability of LVLMs, optimal visual prompts must be discovered through empirical experiments, which have relied on manual human trial-and-error. In this work, we propose an automated semantic exploration framework for discovering task-wise visual prompts. Unlike previous methods, our approach enables diverse yet efficient exploration through agent-driven experiments, minimizing human intervention and avoiding the inefficiency of per-sample generation. We introduce a semantic exploration algorithm named SEVEX, which addresses two major challenges of visual prompt exploration: (1) the distraction caused by lengthy, low-level code and (2) the vast, unstructured search space of visual prompts. Specifically, our method leverages an abstract idea space as a search space, a novelty-guided selection algorithm, and a semantic feedback-driven ideation process to efficiently explore diverse visual prompts based on empirical results. We evaluate SEVEX on the BlindTest and BLINK benchmarks, which are specifically designed to assess LVLM perception. Experimental results demonstrate that SEVEX significantly outperforms baseline methods in task accuracy, inference efficiency, exploration efficiency, and exploration stability. Notably, our framework discovers sophisticated and counterintuitive visual strategies that go beyond conventional tool usage, offering a new paradigm for enhancing LVLM perception through automated, task-wise visual prompts.

ECCV 2026

FaceMoE: Mixture of Experts for Low-Resolution Face Recognition

Kartik Narayan, Vishal Patel

Low-resolution face recognition (LR-FR) remains a challeng-ing task due to poor feature extraction and aggregation, as probe imagesoften contain limited identity information resulting from extreme degrada-tions such as blur, occlusion, and low contrast. Additionally, the domaingap between high-resolution (HR) gallery images and low-resolution (LR)probe images poses a significant challenge. A single feature encoder strug-gles to generalize effectively across both domains when fine-tuned on anLR dataset, and this issue is further magnified by catastrophic forget-ting. To address these challenges, we propose FaceMoE, an effectiveadaptation of Mixture of Experts (MoE) transfomer architecture for low-resolution face-recognition . Specifically, we introduce multiple specializedfeed-forward network (FFN) experts and incorporate a top-k router, whichdynamically assigns tokens to appropriate experts. This design emergentlypromotes specialization across experts for different semantic regions ofthe face, which enables FaceMoE to perform resolution-aware featureextraction. Moreover, the top-k router facilitates sparse expert activation,enabling the model to preserve pretrained knowledge when finetuned on aLR dataset, while increasing model capacity without proportional compu-tational overhead. FaceMoE is trained with a combined face recognitionloss, router z-loss, and load balancing loss to ensure expert specializa-tion and stable training. To the best of our knowledge, this is the firstwork leveraging MoE for LR-FR. Extensive experiments across elevendatasets, spanning HR, mixed-quality, and LR benchmarks, demonstratethat FaceMoE significantly outperforms state-of-the-art methods.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.