论文检索

Sunbeom Jeong, Sehwan Kim, Hyeonggeun Han, Hyungjun Joo, Sangwoo Hong, Jungwoo Lee

Dataset distillation (DD) aims to generate a compact synthetic dataset that enables efficient training of neural networks while maintaining performance comparable to that achieved with the original dataset. However, existing methods often suffer from two main limitations. They either rely on computationally intensive iterative optimization procedures or depend heavily on architecture-specific designs. These issues limit their practicality for large-scale datasets and hinder generalization across different model architectures. To overcome these challenges, recent research has explored the use of diffusion models as an architecture-agnostic approach to dataset distillation, offering improved scalability and generalization for large-scale datasets across diverse model architectures. While diffusion-based dataset distillation methods have shown considerable potential, several challenges remain. Notably, certain approaches exhibit a distributional mismatch between the pre-trained diffusion model and the target dataset, which can adversely affect the fidelity and representativeness of the generated samples. Others require substantial fine-tuning to achieve high fidelity, which negates the benefits of architectural flexibility. In this work, we propose a new diffusion-based dataset distillation framework that effectively preserves the characteristics of the original dataset without requiring any fine-tuning. Our method employs adaptive sampling and repulsion regularization to enhance both the fidelity and diversity of generated samples. As a result, the proposed approach outperforms state-of-the-art distillation methods across a wide range of datasets and model architectures.

Sida Huang, Siqi Huang, Ping Luo, Hongyuan Zhang

With the development of diffusion models, enhancing spatial controllability in text-to-image generation has become a vital challenge. As a representative task for addressing this challenge, layout-to-image generation aims to generate images that are spatially consistent with the given layout condition. Existing layout-to-image methods typically introduce the layout condition by integrating adapter modules into the base generative model. However, the generated images often exhibit low visual quality and stylistic inconsistency with the base model, indicating a loss of pretrained knowledge. To alleviate this issue, we construct the Layout Synthesis (LaySyn) dataset, which leverages images synthesized by the base model itself to mitigate the distribution shift from the pretraining data. Moreover, we propose the Layout Control (Laytrol) Network, in which parameters are inherited from MM-DiT to preserve the pretrained knowledge of the base model. To effectively activate the copied parameters and avoid disturbance from unstable control conditions, we adopt a dedicated initialization scheme for Laytrol. In this scheme, the layout encoder is initialized as a pure text encoder to ensure that its output tokens remain within the data domain of MM-DiT. Meanwhile, the outputs of the layout control network are initialized to zero. In addition, we apply Object-level Rotary Position Embedding to the layout tokens to provide coarse positional information. Qualitative and quantitative experiments demonstrate the effectiveness of our method.

Shaoguang Huang, Yunzhen Wang, Haijin Zeng, Hongyu Chen, Hongyan Zhang

Snapshot compressive imaging (SCI) captures multispectral images (MSIs) using a single coded two-dimensional (2-D) measurement, but reconstructing high-fidelity MSIs from these compressed inputs remains a fundamentally ill-posed challenge. Recent diffusion-based methods improve quality but are limited by scarce MSI training data, domain shifts from RGB-pretrained models, and slow multi-step sampling. These drawbacks restrict their practicality in real-world applications. Unlike prior approaches that rely on expensive iterative refinement or subspace-based diffusion embeddings (e.g., DiffSCI, PSR-SCI)—we introduce a fundamentally different paradigm: a self-supervised One-Step Diffusion (OSD) framework designed specifically for SCI. The key novelty lies in using a single-step diffusion refiner to correct an initial reconstruction, eliminating iterative denoising entirely while preserving generative quality. Moreover, we adopt a self-supervised equivariant learning strategy to train both the predictor and refiner directly from raw 2-D measurements, enabling generalization to unseen domains without ground-truth MSI. To further address limited MSI data, we design a band-selection–driven distillation strategy that transfers core generative priors from large-scale RGB datasets, effectively bridging the domain gap. Extensive experiments confirm that our approach sets a new standard—yielding PSNR gains of 3.44dB, 1.61dB, and 0.28dB on the Harvard, NTIRE, and ICVL datasets respectively, while cutting reconstruction time from 8.9s to just 0.22s per image. These gains in efficiency and adaptability advance SCI reconstruction, enabling accurate and practical real-world deployment.

Xinyuan Hu, Changyue Shi, Chuxiao Yang, Minghao Chen, Jiajun Ding, Tao Wei, Chen Wei, Zhou Yu, Min Tan

Feed-forward 3D reconstruction from sparse, low-resolution (LR) images is a crucial capability for real-world applications, such as autonomous driving and embodied AI. However, existing methods often fail to recover fine texture details. This limitation stems from the inherent lack of high-frequency information in LR inputs. To address this, we propose SRSplat, a feed-forward framework that reconstructs high-resolution 3D scenes from only a few LR views. Our main insight is to compensate for the deficiency of texture information by jointly leveraging external high-quality reference images and internal texture cues. We first construct a scene-specific reference gallery, generated for each scene using Multimodal Large Language Models (MLLMs) and diffusion models. To integrate this external information, we introduce the Reference-Guided Feature Enhancement (RGFE) module, which aligns and fuses features from the LR input images and their reference twin image. Subsequently, we train a decoder to predict the Gaussian primitives using the multi-view fused feature obtained from RGFE. To further refine predicted Gaussian primitives, we introduce Texture-Aware Density Control (TADC), which adaptively adjusts Gaussian density based on the internal texture richness of the LR inputs. Extensive experiments demonstrate that our SRSplat outperforms existing methods on various datasets, including RealEstate10K, ACID, and DTU, and exhibits strong cross-dataset and cross-resolution generalization capabilities.

Teng Hu, Jiangning Zhang, Zihan Su, Ran Yi

Recent advances in video generation have made it possible to produce visually compelling videos, with wide-ranging applications in content creation, entertainment, and virtual reality. However, most existing diffusion transformer based video generation models are limited to low-resolution outputs (

Jichen Hu, Chen Yang, Zanwei Zhou, Jiemin Fang, Qi Tian, Wei Shen

Reflection removal of a single image remains a highly challenging task due to the complex entanglement between target scenes and unwanted reflections. Despite significant progress, existing methods are hindered by the scarcity of high-quality, diverse data and insufficient restoration priors, resulting in limited generalization across various real-world scenarios. In this paper, we propose Dereflection Any Image, a comprehensive solution with an efficient data preparation pipeline and a generalizable model for robust reflection removal. First, we introduce a dataset named Diverse Reflection Removal (DRR) created by randomly rotating reflective mediums in target scenes, enabling variation of reflection angles and intensities, and setting a new benchmark in scale, quality, and diversity. Second, we propose a diffusion-based framework with one-step diffusion for deterministic outputs and fast inference. To ensure stable learning, we design a three-stage progressive training strategy, including reflection-invariant finetuning to encourage consistent outputs across varying reflection patterns that characterize our dataset. Extensive experiments show that our method achieves SOTA performance on both common benchmarks and challenging in-the-wild images, showing superior generalization across diverse real-world scenes.

Chenggong Hu, Yi Wang, Mengqi Xue, Haofei Zhang, Jie Song, Li Sun

Textile pattern generation (TPG) aims to synthesize fine-grained textile pattern images based on given clothing images. Although previous studies have not explicitly investigated TPG, existing image-to-image models appear to be natural candidates for this task. However, when applied directly, these methods often produce unfaithful results, failing to preserve fine-grained details due to feature confusion between complex textile patterns and the inherent non-rigid texture distortions in clothing images. In this paper, we propose a novel method, SLDDM-TPG, for faithful and high-fidelity TPG. Our method consists of two stages: (1) a latent disentangled network (LDN) that resolves feature confusion in clothing representations and constructs a multi-dimensional, independent clothing feature space; and (2) a semi-supervised latent diffusion model (S-LDM), which receives guidance signals from LDN and generates faithful results through semi-supervised diffusion training, combined with our designed fine-grained alignment strategy. Extensive evaluations show that SLDDM-TPG reduces FID by 4.1 and improves SSIM by up to 0.116 on our CTP-HD dataset, and also demonstrate good generalization on the VITON-HD dataset.

Liang Hou, Cong Liu, Mingwu Zheng, Xin Tao, Pengfei Wan, Di Zhang, Kun Gai

Resolution generalization in image generation tasks enables the production of higher-resolution images with lower training resolution overhead. However, a key obstacle for diffusion transformers in addressing this problem is the mismatch between positional encodings seen at inference and those used during training. Existing strategies such as positional encodings interpolation, extrapolation, or hybrids, do not fully resolve this mismatch. In this paper, we propose a novel two-dimensional randomized positional encodings, namely RPE-2D, that prioritizes the order of image patches rather than their absolute distances, enabling seamless high- and low-resolution generation without training on multiple resolutions. Concretely, RPE-2D independently samples positions along the horizontal and vertical axes over an expanded range during training, ensuring that the encodings used at inference lie within the training distribution and thereby improving resolution generalization. We further introduce a simple random resize-and-crop augmentation to strengthen order modeling and add micro-conditioning to indicate the applied cropping pattern. On the ImageNet dataset, RPE-2D achieves state-of-the-art resolution generalization performance, outperforming competitive methods when trained at 256^2 and evaluated at 384^2 and 512^2, and when trained at 512^2 and evaluated at 768^2 and 1024^2. RPE-2D also exhibits outstanding capabilities in low-resolution image generation, multi-stage training acceleration, and multi-resolution inheritance.

Junming Hou, Ran Ran, Sixing Chen, Zihao Chen, Xiaofeng Cong, Junling Li, Liang-Jian Deng

Pansharpening is a powerful technique for generating high-resolution multispectral (HRMS) images by fusing currently available image pairs of low-resolution multispectral (LRMS) and texture-rich panchromatic (PAN) data, effectively addressing the physical constraints of satellite sensors. While recent generative diffusion models have demonstrated impressive performance gains in this domain, their prohibitive computational demands and training costs hinder practicality in resource-constrained remote sensing satellite systems. In this work, we propose NODiff, a novel diffusion framework that replaces the conventional attention-based denoising backbone with a neural operator, seamlessly integrating operator learning and generative modeling into an efficient yet effective solution for pansharpening. In practice, we implement our approach through a two-stage learning paradigm: First, we pretrain the proposed Neural Operator-based diffusion model to learn the high-resolution texture priors essential for pansharpening. Afterward, we freeze the pretrained parameters, and design a lightweight conditional detail guidance adapter to enable efficient fine-tuning for generating desired HRMS images. Meanwhile, a time-aware low-rank adaptation is introduced to dynamically refine high-frequency details potentially affected by spectral mode truncation. Extensive experiments on multiple benchmark datasets demonstrate that NODiff achieves competitive pansharpening performance while significantly reducing training and inference costs. Beyond pansharpening, our method provides new insights into building resource-efficient generative models.

Jiwoo Hong, Sayak Paul, Noah Lee, Kashif Rasul, James Thorne, Jongheon Jeong

Modern preference alignment methods, such as DPO, rely on divergence regularization to a reference model for training stability—but this creates a fundamental problem we call "reference mismatch." In this paper, we investigate the negative impacts of reference mismatch in aligning text-to-image (T2I) diffusion models, showing that larger reference mismatch hinders effective adaptation given the same amount of data, e.g., as when learning new artistic styles, or personalizing to specific objects. We demonstrate this phenomenon across text-to-image (T2I) diffusion models and introduce margin-aware preference optimization (MaPO), a reference-agnostic approach that breaks free from this constraint. By directly optimizing the likelihood margin between preferred and dispreferred outputs under the Bradley-Terry model without anchoring to a reference, MaPO transforms diverse T2I tasks into unified pairwise preference optimization. We validate MaPO's versatility across five challenging domains: (1) safe generation, (2) style adaptation, (3) cultural representation, (4) personalization, and (5) general preference alignment. Our results reveal that MaPO's advantage grows dramatically with reference mismatch severity, outperforming both DPO and specialized methods like DreamBooth while reducing training time by 15%. MaPO thus emerges as a versatile and memory-efficient method for generic T2I adaptation tasks.

Xiao He, Zhijun Tu, Kun Cheng, Mingrui Zhu, Jie Hu, Nannan Wang, Xinbo Gao

The demonstrated success of sparsely-gated Mixture-of-Experts (MoE) architectures, exemplified by models such as DeepSeek and Grok, has motivated researchers to investigate their adaptation to diverse domains. In real-world image super-resolution (Real-ISR), existing approaches mainly rely on fine-tuning pre-trained diffusion models through Low-Rank Adaptation (LoRA) module to reconstruct high-resolution (HR) images. However, these dense Real-ISR models are limited in their ability to adaptively capture the heterogeneous characteristics of complex real-world degraded samples or enable knowledge sharing between inputs under equivalent computational budgets. To address this, we investigate the integration of sparse MoE into Real-ISR and propose a Mixture-of-Ranks (MoR) architecture for single-step image super-resolution. We introduce a fine-grained expert partitioning strategy that treats each rank in LoRA as an independent expert. This design enables flexible knowledge recombination while isolating fixed-position ranks as shared experts to preserve common-sense features and minimize routing redundancy. Furthermore, we develop a degradation estimation module leveraging CLIP embeddings and predefined positive-negative text pairs to compute relative degradation scores, dynamically guiding expert activation. To better accommodate varying sample complexities, we incorporate zero-expert slots and propose a degradation-aware load-balancing loss, which dynamically adjusts the number of active experts based on degradation severity, ensuring optimal computational resource allocation. Comprehensive experiments validate our framework's effectiveness and state-of-the-art performance.

Wenzhe He, Xiaojun Chen, Ruiqi Wang, Ruihui Li, Huilong Pi, Jiapeng Zhang, Zhuo Tang, Kenli Li

3D LiDAR scene completion from point clouds is a fundamental component of perception systems in autonomous vehicles. Previous methods have predominantly employed diffusion models for high‑fidelity reconstruction. However, their multi-step iterative sampling incurs significant computational overhead, limiting its real-time applicability. To address this, we propose LiNeXt: a lightweight, non‐diffusion network optimized for rapid and accurate point cloud completion. Specifically, LiNeXt first applies the Noise‑to‑Coarse (N2C) Module to denoise the input noisy point cloud in a single pass, thereby obviating the multi‑step iterative sampling of diffusion‑based methods. The Refine Module then takes the coarse point cloud and its intermediate features from the N2C Module to perform more precise refinement, further enhancing structural completeness. Furthermore, we observe that LiDAR point clouds exhibit a distance-dependent spatial distribution, being densely sampled at proximal ranges and sparsely sampled at distal ranges. Accordingly, we propose the Distance‑aware Selected Repeat strategy to generate a more uniformly distributed noisy point cloud. On the SemanticKITTI dataset, LiNeXt achieves a 199.8 times speedup in inference, reduces Chamfer Distance by 50.7 percent, and uses only 6.1 percent of the parameters compared with LiDiff. These results demonstrate the superior efficiency and effectiveness of LiNeXt for real-time scene completion.

Haoran He, Yang Zhang, Liang Lin, Zhongwen Xu, Ling Pan

Video generative models pre-trained on large-scale internet datasets have achieved remarkable success, excelling at producing realistic synthetic videos. However, they often generate clips based on static prompts (e.g., text or images), limiting their ability to model interactive and dynamic scenarios. In this paper, we propose Dynamic World Simulation (DWS), a novel approach to transform pre-trained video generative models into controllable world simulators capable of executing specified action trajectories. To achieve precise alignment between conditioned actions and generated visual changes, we introduce a lightweight, universal action-conditioned module that seamlessly integrates into any existing model. Instead of focusing on complex visual details, we demonstrate that consistent dynamic transition modeling is the key to building powerful world simulators. Building upon this insight, we further introduce a motion-reinforced loss that enhances action controllability by compelling the model to capture dynamic changes more effectively. Experiments demonstrate that DWS can be versatilely applied to both diffusion and autoregressive transformer models, achieving significant improvements in generating action-controllable, dynamically consistent videos across games and robotics domains. Moreover, to facilitate the applications of the learned world simulator in downstream tasks such as model-based reinforcement learning, we propose prioritized imagination to improve sample efficiency, demonstrating competitive performance compared with state-of-the-art methods.

Chen Hang, Haoming Chen, Xuwei Fang, Weisheng Xie, Xiangxiang Gao, Faming Fang, Guixu Zhang, Haichuan Song

In this paper, we investigate the limitations of the Vector Quantized Latent Diffusion Model (VQ-LDM) in restoration tasks. We identify a performance gap between the Vector Quantization (VQ) and Diffusion Model components, manifested as a significant discrepancy between the reconstruction quality of ground truth images processed via VQ autoregression and degraded images restored by VQ-LDM. Through experiments, we attribute this gap primarily to the lack of robustness in the mapped points of VQ within the original VQ-LDM framework. To address this issue, we propose a geometric based optimization approach. First, we introduce a simple yet effective method, termed interpolation-based latent initial state optimization, which mitigates the performance gap by replacing the original mapped points with interpolated values, supported by theoretical analysis. Here, the latent initial state refers specifically to the input of the diffusion model. Building upon this, we further propose a Chebyshev center-based latent initial state optimization, an elegant theoretical solution from a geometric perspective, that further enhances restoration performance. Our improvements consistently achieve superior results across nine benchmark datasets.

Dong Han, Yong Li, Joachim Denzler

With the advancement of face recognition (FR) systems, privacy-preserving face recognition (PPFR) systems have gained popularity for their accurate recognition, enhanced facial privacy protection, and robustness to various attacks. However, there are limited studies to further verify privacy risks by reconstructing realistic high-resolution face images from embeddings of these systems, especially for PPFR. In this work, we propose the face embedding mapping (FEM), a general framework that explores Kolmogorov-Arnold Network (KAN) for conducting the embedding-to-face attack by leveraging pre-trained Identity-Preserving diffusion model against state-of-the-art (SOTA) FR and PPFR systems. Based on extensive experiments, we verify that reconstructed faces can be used for accessing other real-word FR systems. Besides, the proposed method shows the robustness in reconstructing faces from the partial and protected face embeddings. Moreover, FEM can be utilized as a tool for evaluating safety of FR and PPFR systems in terms of privacy leakage. All images used in this work are from public datasets.

Yiyang Gu, Bohan Wu, Yifang Qin, Jiaru Tang, Rong-Cheng Tu, Zhiping Xiao, Taian Guo, Junyu Luo, Wei Ju, Xiao Luo 等

In this paper, we study the problem of robust multi-label hashing, where label noise hinders the learning of a reliable semantic structure from data. Many existing methods rely on heuristic sample selection or consistency-based training, but lack a unified mechanism to validate and refine supervision across structural and semantic levels. Inspired by cognitive theories of human memory, we propose a novel framework called CogniTrust that unifies verifiable supervision with a triadic memory model: a) In episodic memory, feature activations are decomposed into spatial patterns that support the assessment of structural evidence and the estimation of label reliability; b) Semantic memory keeps track of class-level prototypes from structurally attentive regions to estimate the semantic plausibility of labels; c) Reconstructive memory simulates memory recall through interpolation between images using a diffusion-based mixup process, which enriches the training signals for semantically uncertain regions. These components work together, allowing supervision to be refined through the joint consideration of spatial structure and semantic information. Extensive experiments on noisy hashing benchmarks demonstrate that CogniTrust consistently outperforms a range of state-of-the-art baselines. Our results show that cognitive memory mechanisms offer a principled basis for more reliable label denoising and robust hashing.

Paul Grimal, Michael Soumm, Hervé Le Borgne, Olivier Ferret, Akihiro Sugimoto

State-of-the-art text-to-image models produce visually impressive results but often struggle with precise alignment to text prompts, leading to missing critical elements or unintended blending of distinct concepts. We propose a novel approach that learns a high-success-rate distribution conditioned on a target prompt, ensuring that generated images faithfully reflect the corresponding prompts. Our method explicitly models the signal component during the denoising process, offering fine-grained control that mitigates over-optimization and out-of-distribution artifacts. Moreover, our framework is training-free and seamlessly integrates with both existing diffusion and flow matching architectures. It also supports additional conditioning modalities -- such as bounding boxes -- for enhanced spatial alignment. Extensive experiments demonstrate that our approach outperforms current state-of-the-art methods.

Kexin Gong, Puyi Yao, Guiyang Luo, Quan Yuan, Tiange Fu, Hui Zhang, Jinglin Li

Collaborative perception leveraging intermediate feature fusion has emerged as a leading paradigm to significantly enhance the environmental perception capabilities of autonomous driving systems. However, existing methods typically rely on discriminative supervision guided by downstream tasks. This paradigm compels models to learn minimal, task-specific representations, which conflicts with the goal of cooperative perception to capture comprehensive information, thereby limiting generalization. To address this issue, we propose DiGS-CP, a novel two-stage generative supervised collaborative perception framework. Specifically, we introduce a diffusion-based generative task that conditions on fused object-level features to generate representations of object-level point clouds. The proposed generative supervision provides fine-grained, task-agnostic signals that encourages the fusion module to learn comprehensive representations beyond task-specific requirements. By preserving and integrating complementary information from collaborative agents, our approach overcomes the limitations of task-specific learning and enhances the generalizability of the learned features. Furthermore, our two-stage architecture requires agents to transmit only object-level features, significantly reducing communication overhead. Extensive experiments on three benchmark datasets demonstrate that DiGS-CP achieves state-of-the-art performance in 3D object detection, while maintaining low bandwidth requirements and exhibiting excellent generalization ability.

Chao Gong, Dong Li, Yingwei Pan, Jingjing Chen, Ting Yao, Tao Mei

Text-guided image inpainting endeavors to generate new content within specified regions of images using textual prompts from users. The primary challenge is to accurately align the inpainted areas with the user-provided prompts while maintaining a high degree of visual fidelity. While existing inpainting methods have produced visually convincing results by leveraging the pre-trained text-to-image diffusion models, they still struggle to uphold both prompt alignment and visual rationality simultaneously. In this work, we introduce FreeInpaint, a plug-and-play tuning-free approach that directly optimizes the diffusion latents on the fly during inference to improve the faithfulness of the generated images. Technically, we introduce a prior-guided noise optimization method that steers model attention towards valid inpainting regions by optimizing the initial noise. Furthermore, we meticulously design a composite guidance objective tailored specifically for the inpainting task. This objective efficiently directs the denoising process, enhancing prompt alignment and visual rationality by optimizing intermediate latents at each step. Through extensive experiments involving various inpainting diffusion models and evaluation metrics, we demonstrate the effectiveness and robustness of our proposed FreeInpaint.

Yujie Gao, ChenCheng Wang, Xianbing Sun, Jiahui Zhan, Wentao Wang, Yiyi Zhang, Haohua Zhao, Liqing Zhang, Jianfu Zhang

In this work, we introduce a novel high-fidelity full-head 3D avatar generation method from a single image, regardless of perspective, style, expression, or accessories. Prior works often fail to preserve consistent head geometry and facial details, primarily due to their limited capacity in modeling fine-grained facial textures and maintaining identity information. To address these challenges, we construct a new high-quality dataset containing 227 sequences of digital human portraits captured from 96 different perspectives, totalling 21,792 frames, featuring high-quality facial texture details. To further improve performance, we propose a novel multi-view diffusion named ID-TS diffusion model, which integrate identity and expression information into the two-stage multi-view diffusion process. The low-resolution stage ensures structural consistency of heads across multiple views, while the high-resolution stage preserves facial detail fidelity and coherence. Finally, we propose an enhanced feed-forward Gaussian avatar reconstruction method that optimizes the network on multi-view images of each single subject, significantly improving 3D facial texture details. Extensive experiments show that our method demonstrates robust performance across challenging scenarios, while showcasing broad applicability across numerous downstream tasks.