Recent studies in extreme image compression have achieved remarkable performance by compressing the tokens from generative tokenizers. However, these methods often prioritize clustering common semantics within the dataset, while overlooking the diverse details of individual objects. Consequently, this results in suboptimal reconstruction fidelity, especially at low bitrates. To address this issue, we introduce a Dual-generative Latent Fusion (DLF) paradigm. DLF decomposes the latent into semantic and detail elements, compressing them through two distinct branches. The semantic branch clusters high-level information into compact tokens, while the detail branch encodes perceptually critical details to enhance the overall fidelity. Additionally, we propose a cross-branch interactive design to reduce redundancy between the two branches, thereby minimizing the overall bit cost. Experimental results demonstrate the impressive reconstruction quality of DLF even below 0.01 bits per pixel (bpp). On the CLIC2020 test set, our method achieves bitrate savings of up to 27.93% on LPIPS and 53.55% on DISTS compared to MS-ILLM. Furthermore, DLF surpasses recent diffusion-based codecs in visual fidelity while maintaining a comparable level of generative realism. Project: https://dlfcodec.github.io/
论文检索
输入标题、作者或关键词，从 2,701 篇学术成果中精准定位
Machine Unlearning (MU) aims to selectively erase harmful behaviors from models while retaining the overall utility of the model. As a multi-task learning problem, MU involves balancing objectives related to forgetting specific concepts/data and preserving general performance. A naive integration of these forgetting and preserving objectives can lead to gradient conflicts and dominance, impeding MU algorithms from reaching optimal solutions.To address the gradient conflict and dominance issue, we reformulate MU as a two-player cooperative game, where the two players, namely, the forgetting player and the preservation player, contribute via their gradient proposals to maximize their overall gain and balance their contributions.To this end, inspired by the Nash bargaining theory, we derive a closed-form solution to guide the model toward the Pareto stationary point.Our formulation of MU guarantees an equilibrium solution, where any deviation from the final state would lead to a reduction in the overall objectives for both players, ensuring optimality in each objective.We evaluate our algorithm's effectiveness on a diverse set of tasks across image classification and image generation.Extensive experiments with ResNet, vision-language model CLIP, and text-to-image diffusion models demonstrate that our method outperforms state-of-the-art MU algorithms, achieving a better trade-off between forgetting and preserving.Our results also highlight improvements in forgetting precision, preservation of generalization, and robustness against adversarial attacks.
Deep learning (DL) has shown transformative potential across industries, yet its sensitivity to adversarial examples (AEs) limits its reliability and broader deployment. Research on DL robustness has developed various techniques, with adversarial training (AT) established as a leading approach to counter AEs. Traditional AT focuses on worst-case robustness (WCR), but recent work has introduced probabilistic robustness (PR), which evaluates the proportion of AEs within a local perturbation range, providing an overall assessment of the model's local robustness and acknowledging residual risks that are more practical to manage. However, existing AT methods are fundamentally designed to improve WCR, and no dedicated methods currently target PR. To bridge this gap, we reformulate a new min-max optimization as the theoretical foundation for AT focused on PR, and introduce an AT-PR training scheme with effective numerical algorithms to solve the new optimization problem. Our experiments, based on 38 DL models trained on common datasets and architectures, demonstrate that AT-PR achieves higher improvements in PR than AT-WCR methods and shows more consistent effectiveness across varying local inputs, with a smaller trade-off in model generalization. Open-source tools and all experiments are publicly accessible.
Learning to forecast trajectories of intelligent agents has caught much more attention recently. However, it remains a challenge to accurately account for agents' intentions and social behaviors when forecasting, and in particular, to simulate the unique randomness within each of those components in an explainable and decoupled way. Inspired by vibration systems and their resonance properties, we propose the Resonance (short for Re) model to encode and forecast pedestrian trajectories in the form of "co-vibrations". It decomposes trajectory modifications and randomness into multiple vibration portions to simulate agents' reactions to each single cause, and forecasts trajectories as the superposition of these independent vibrations separately. Also, benefiting from such vibrations and their spectral properties, representations of social interactions can be learned by emulating the resonance phenomena, further enhancing its explainability. Experiments on multiple datasets have verified its usefulness both quantitatively and qualitatively.
We propose SynCity, a method for generating explorable 3D worlds from textual descriptions. Our approach leverages pre-trained textual, image, and 3D generators without requiring fine-tuning or inference-time optimization. While most 3D generators are object-centric and unable to create large-scale worlds, we demonstrate how 2D and 3D generators can be combined to produce ever-expanding scenes. The world is generated tile by tile, with each new tile created within its context and seamlessly integrated into the scene. SynCity enables fine-grained control over the appearance and layout of the generated worlds, which are both detailed and diverse.
Sequential keypoint density estimator: an overlooked baseline of skeleton-based video anomaly detection
PDF ↗Detecting anomalous human behaviouris an important visual taskin safety-critical applicationssuch as healthcare monitoring,workplace safety,or public surveillance.In these contexts,abnormalities are often reflectedwith unusual human poses.Thus, we propose SeeKer,a method for detecting anomaliesin sequences of human skeletons.Our method formulates the skeleton sequence densitythrough autoregressive factorization at the keypoint level.The corresponding conditional distributionsrepresent probable keypoint locations given prior skeletal motion.We formulate the joint distribution of the considered skeletonas causal prediction of conditional Gaussiansacross its constituent keypoints.A skeleton is flagged as anomalous if its keypoint locations surprise our model(i.e. receive a low density).In practice, our anomaly score is a weighted sum of per-keypoint log-conditionals,where the weights account for the confidence of the underlying keypoint detector.Despite its conceptual simplicity,SeeKer surpasses all previous methodson the UBnormal and MSAD-HR datasetswhile delivering competitive performanceon the ShanghaiTech dataset.
EmbodiedSplat: Personalized Real-to-Sim-to-Real Navigation with Gaussian Splats from a Mobile Device
PDF ↗The field of Embodied AI predominantly relies on simulation for training and evaluation, often using either fully synthetic environments that lack photorealism or high-fidelity real-world reconstructions captured with expensive hardware. As a result, sim-to-real transfer remains a major challenge. In this paper, we introduce EmbodiedSplat, a novel approach that personalizes policy training by efficiently capturing the deployment environment and fine-tuning policies within the reconstructed scenes. Our method leverages 3D Gaussian Splatting (GS) and the Habitat-Sim simulator to bridge the gap between realistic scene capture and effective training environments. Using iPhone-captured deployment scenes, we reconstruct meshes via GS, enabling training in settings that closely approximate real-world conditions. We conduct a comprehensive analysis of training strategies, pre-training datasets, and mesh reconstruction techniques, evaluating their impact on sim-to-real predictivity in real-world scenarios. Experimental results demonstrate that agents fine-tuned with EmbodiedSplat outperform both zero-shot baselines pre-trained on large-scale real-world datasets (HM3D) and synthetically generated datasets (HSSD), achieving absolute success rate improvements of 20% and 40% on real-world Image Navigation task. Moreover, our approach yields a high sim-vs-real correlation (0.87-0.97) for the reconstructed meshes, underscoring its effectiveness in adapting policies to diverse environments with minimal effort. Project page: https://gchhablani.github.io/embodied-splat.
Recent studies have shown that deep learning models are vulnerable to membership inference attacks (MIAs), which aim to infer whether a data record was used to train a target model or not. To analyze and study these vulnerabilities, various MIA methods have been proposed. Despite the significance and popularity of MIAs, existing works on MIAs are limited in providing guarantees on the false discovery rate (FDR), which refers to the expected proportion of false discoveries among the identified positive discoveries. However, it is very challenging to ensure the false discovery rate guarantees, because the underlying distribution is usually unknown, and the estimated non-member probabilities often exhibit interdependence. To tackle the above challenges, in this paper, we design a novel membership inference attack method, which can provide the guarantees on the false discovery rate. Additionally, we show that our method can also provide the marginal probability guarantee on labeling true non-member data as member data. Notably, our method can work as a wrapper that can be seamlessly integrated with existing MIA methods in a post-hoc manner, while also providing the FDR control. We perform the theoretical analysis for our method. Extensive experiments in various settings (e.g., the black-box setting and the lifelong learning setting) are also conducted to verify the desirable performance of our method.
With the rapid advancement of diffusion-based generative models, portrait image animation has achieved remarkable results. However, it still faces challenges in temporally consistent video generation and fast sampling due to its iterative sampling nature. This paper presents FLOAT, an audio-driven talking portrait video generation method based on flow matching generative model. Instead of a pixel-based latent space, we take advantage of a learned orthogonal motion latent space, enabling efficient generation and editing of temporally consistent motion. To achieve this, we introduce a transformer-based vector field predictor with an effective frame-wise conditioning mechanism. Additionally, our method supports speech-driven emotion enhancement, enabling a natural incorporation of expressive motions. Extensive experiments demonstrate that our method outperforms state-of-the-art audio-driven talking portrait methods in terms of visual quality, motion fidelity, and efficiency.
The pretrain-finetune paradigm of pre-training a model on large amounts of image and text data and then fine-tuning the model for a specific task has led to significant progress in many 2D image and natural language processing tasks.Similarly, the use of pre-training methods in point cloud data can also enhance the working performance and generalization ability of the model.Therefore, in this paper, we propose a pre-training framework based on a diffusion model called PreDifPoint. It is able to accomplish the pre-training of the model's backbone network through a diffusion process of gradual denoising. We aggregate the potential features extracted from the backbone network, input them as conditions into the subsequent diffusion model, and direct the point-to-point mapping relationship of the noisy point clouds at neighboring time steps, so as to generate high-quality point clouds and at the same time better perform various downstream tasks of the point clouds.We also introduce a bi-directional covariate attention (DXCA-Attention) mechanism for capturing complex feature interactions, fusing local and global features, and improving the detail recovery of point clouds.In addition, we propose a density-adaptive sampling strategy, which can help the model dynamically adjust the sampling strategy between different time steps, and guide the model to pay more attention to the denser regions in the point cloud, thus improving the effectiveness of the model in point cloud recovery.Our PreDifPoint framework achieves more competitive results on various real-world datasets. Specifically, PreDifPoint achieves an overall accuracy of 87.96%, which is 0.35% higher than PointDif, on the classification task on PB-T50-395RS, a variant of ScanObjectNN dataset.
Towards Stabilized and Efficient Diffusion Transformers through Long-Skip-Connections with Spectral Constraints
PDF ↗Diffusion Transformers (DiT) have emerged as a powerful architecture for image and video generation, offering superior quality and scalability. However, their practical application suffers from inherent dynamic feature instability, leading to error amplification during cached inference. Through systematic analysis, we identify the absence of long-range feature preservation mechanisms as the root cause of unstable feature propagation and perturbation sensitivity. To this end, we propose Skip-DiT, a novel DiT variant enhanced with Long-Skip-Connections (LSCs) - the key efficiency component in U-Nets. Theoretical spectral norm and visualization analysis demonstrate how LSCs stabilize feature dynamics. Skip-DiT architecture and its stabilized dynamic feature enable an efficient statical caching mechanism that reuses deep features across timesteps while updating shallow components. Extensive experiments across image and video generation tasks demonstrate that Skip-DiT achieves: (1) **4.4x** training acceleration and faster convergence, (2) **1.5-2x** inference acceleration without quality loss and high fidelity to original output, outperforming existing DiT caching methods across various quantitative metrics. Our findings establish long-skip connections as critical architectural components for training stable and efficient diffusion transformers. Codes are provided in the anonymous URL https://anonymous.4open.science/r/Skip-DiT-72B7/.
TAB: Transformer Attention Bottlenecks enable User Intervention and Debugging in Vision-Language Models
PDF ↗Multi-head self-attention (MHSA) is a key component of Transformers, a widely popular architecture in both language and vision. Multiple heads intuitively enable different parallel processes over the same input. Yet, they also obscure the attribution of each input patch to the output of a model. We propose a novel 1-head Transformer Attention Bottleneck (TAB) layer, inserted after the traditional MHSA architecture, to serve as an attention bottleneck for interpretability and intervention. Unlike standard self-attention, TAB constrains the total attention over all patches to \in [0, 1]. That is, when the total attention is 0, no visual information is propagated further into the network, and the vision-language model (VLM) would default to a generic, image-independent response. To demonstrate the advantages of TAB, we train VLMs with TAB to perform image-difference captioning. Over three datasets, our models perform similarly to baseline VLMs in captioning but the bottleneck is superior in localizing changes and in identifying when no changes occur. TAB is the first architecture to enable users to debug by editing attention, which often produces expected outputs by VLMs.
DropletVideo: A Dataset and Approach to Explore Integral Spatio-Temporal Consistent Video Generation
PDF ↗Spatio-temporal consistency is a critical topic in video generation. A qualified generated video segment must ensure plot plausibility and coherence while maintaining visual consistency of objects and scenes across varying viewpoints. Prior research, especially in open-source projects, primarily focuses on either temporal or spatial consistency, or their basic combination, such as appending a camera-movement description after a prompt without constraining its outcomes. However, camera movement may introduce new objects to the scene or eliminate existing ones, thereby affecting the preceding narrative. Especially in videos with numerous camera movements, the interplay between multiple plots becomes increasingly complex. This paper introduces and examines integral spatio-temporal consistency, considering the synergy between plot progression and camera techniques, and the long-term impact of prior content on subsequent generation. Our research encompasses dataset construction through to model development. Initially, we constructed DropletVideo-10M, which comprises 10 million videos that feature dynamic camera motion and object actions. With an average length of 206 words, the captions offer detailed accounts of camera movements. Following this, we developed the DropletVideo model, which excels in preserving spatio-temporal coherence during video generation. The work has been open-sourced: https://dropletx.github.io/.
Relational video customization refers to the creation of personalized videos that depict user-specified relations between two subjects, a crucial task for comprehending real-world visual content. While existing methods can personalize subject appearances and motions, they still struggle with complex relational video customization, where precise relational modeling and high generalization across subject categories are essential. The primary challenge arises from the intricate spatial arrangements, layout variations, and nuanced temporal dynamics inherent in relations; consequently, current models tend to overemphasize irrelevant visual details rather than capturing meaningful interactions. To address these challenges, we propose DreamRelation, a novel approach that personalizes relations through a small set of exemplar videos, leveraging two key components: Relational Decoupling Learning and Relational Dynamics Enhancement. First, in Relational Decoupling Learning, we disentangle relations from subject appearances using relation LoRA triplet and hybrid mask training strategy, ensuring better generalization across diverse relationships. Furthermore, we determine the optimal design of relation LoRA triplet by analyzing the distinct roles of the query, key, and value features within MM-DiT's attention mechanism, making DreamRelation the first relational video generation framework with explainable components. Second, in Relational Dynamics Enhancement, we introduce space-time relational contrastive loss, which prioritizes relational dynamics while minimizing the reliance on detailed subject appearances. Extensive experiments demonstrate that DreamRelation outperforms state-of-the-art methods in relational video customization. Project page at: https://dreamrelation.github.io/.
We present SAM4D, a multi-modal and temporal foundation model designed for promptable segmentation across camera and LiDAR streams. Unified Multi-modal Positional Encoding (UMPE) is introduced to align camera and LiDAR features in a shared 3D space, enabling seamless cross-modal prompting and interaction. Additionally, we propose Motion-aware Cross-modal Memory Attention (MCMA), which leverages ego-motion compensation to enhance temporal consistency and long-horizon feature retrieval, ensuring robust segmentation across dynamically changing autonomous driving scenes. To avoid annotation bottlenecks, we develop a multi-modal automated data engine that synergizes VFM-driven video masklets, spatiotemporal 4D reconstruction, and cross-modal masklet fusion. This framework generates camera-LiDAR aligned pseudo-labels at a speed orders of magnitude faster than human annotation while preserving VFM-derived semantic fidelity in point cloud representations. We conduct extensive experiments on the constructed Waymo-4DSeg, which demonstrate the powerful cross-modal segmentation ability and great potential in data annotation of proposed SAM4D.
VLIPP: Towards Physically Plausible Video Generation with Vision and Language Informed Physical Prior
PDF ↗Video diffusion models (VDMs) have advanced significantly in recent years, enabling the generation of highly realistic videos and drawing the attention of the community in their potential as world simulators. However, despite their capabilities, VDMs often fail to produce physically plausible videos due to an inherent lack of understanding of physics, resulting in incorrect dynamics and event sequences. To address this limitation, we propose a novel two-stage image-to-video generation framework that explicitly incorporates physics with vision and language informed physical prior. In the first stage, we employ a Vision Language Model (VLM) as a coarse-grained motion planner, integrating chain-of-thought and physics-aware reasoning to predict a rough motion trajectories/changes that approximate real-world physical dynamics while ensuring the inter-frame consistency. In the second stage, we use the predicted motion trajectories/changes to guide the video generation of a VDM. As the predicted motion trajectories/changes are rough, noise is added during inference to provide freedom to the VDM in generating motion with more fine details. Extensive experimental results demonstrate that our framework can produce physically plausible motion, and comparative evaluations highlight the notable superiority of our approach over existing methods. More video results and code are available on our Project Page: https://madaoer.github.io/projects/physically_plausible_video_generation/.
LiON-LoRA: Rethinking LoRA Fusion to Unify Controllable Spatial and Temporal Generation for Video Diffusion
PDF ↗Video Diffusion Models (VDMs) have demonstrated remarkable capabilities in synthesizing realistic videos by learning from large-scale data. Although vanilla Low-Rank Adaptation (LoRA) can learn specific spatial or temporal movement to driven VDMs with constrained data, achieving precise control over both camera trajectories and object motion remains challenging due to the unstable fusion and non-linear scalability. To address these issues, we propose LiON-LoRA, a novel framework that rethinks LoRA fusion through three core principles: Linear scalability, Orthogonality, and Norm consistency. First, we analyze the orthogonality of LoRA features in shallow VDM layers, enabling decoupled low-level controllability. Second, norm consistency is enforced across layers to stabilize fusion during complex camera motion combinations. Third, a controllable token is integrated into the diffusion transformer (DiT) to linearly adjust motion amplitudes for both cameras and objects with a modified self-attention mechanism to ensure decoupled control. Additionally, we extend LiON-LoRA to temporal generation by leveraging static-camera videos, unifying spatial and temporal controllability. Experiments demonstrate that LiON-LoRA outperforms state-of-the-art methods in trajectory control accuracy and motion strength adjustment, achieving superior generalization with minimal training data.
We introduce ArtEditor, a novel framework for instruction-based image editing that learns unique editing styles from few-shot examples. While image editing has seen significant advancements, customized instructional editing remains underexplored. Existing methods often rely on complex, multi-stage pipelines that are difficult to adapt to specific styles. Additionally, this domain lacks a standardized benchmark, making it challenging to evaluate progress. To address these issues, we propose ArtEditor, a two-stage training framework. In the first stage, we train ArtEditor-Base, a general-purpose image editing model, on large-scale datasets to build a strong foundational capability. In the second stage, we fine-tune this model using ArtEditor-LoRA, a lightweight adaptation module, on a small dataset of before-and-after image pairs. This approach enables the model to efficiently learn distinct editing styles and techniques with minimal data. To enhance the performance of a pre-trained Diffusion Transformer (DiT) model, we introduce two key innovations: position encoding cloning and a noise-free conditioning paradigm. These techniques ensure stable and coherent edits, even when adapting to new styles. To support research in this area, we contribute the DoodleArt dataset, the first benchmark specifically designed for customized image editing. DoodleArt features six high-quality artistic styles created by professional artists and designers, providing a valuable resource for evaluating and advancing future work. Extensive experiments demonstrate that ArtEditor achieves superior performance and robustness in customized image editing. Our framework opens new possibilities for artistic creation, offering artists intuitive and flexible tools to bring their visions to life.
The integration of geometric reconstruction and generative modeling remains a critical challenge in developing AI systems capable of human-like spatial reasoning. This paper proposes Aether, a unified framework that enables geometry-aware reasoning in world models by jointly optimizing three core capabilities: (1) 4D dynamic reconstruction, (2) action-conditioned video prediction, and (3) goal-conditioned visual planning. Through task-interleaved feature learning, Aether achieves synergistic knowledge sharing across reconstruction, prediction, and planning objectives. Building upon video generation models, our framework demonstrates zero-shot synthetic-to-real generalization despite never observing real-world data during training. Furthermore, our approach achieves zero-shot generalization in both action following and reconstruction tasks, thanks to its intrinsic geometric modeling. Notably, even without real-world data, its reconstruction performance is comparable with or even better than that of domain-specific models. Additionally, Aether employs camera trajectories as geometry-informed action spaces, enabling effective action-conditioned prediction and visual planning. We hope our work inspires the community to explore new frontiers in physically-reasonable world modeling and its applications.
Vision Transformers (ViTs) have emerged as a foundational model in computer vision, excelling in generalization and adaptation to downstream tasks. However, deploying ViTs to support diverse resource constraints typically requires retraining multiple, size-specific ViTs, which is both time-consuming and energy-intensive. To address this issue, we propose an efficient ViT adaptation framework that enables a single adaptation process to generate multiple models of varying sizes for deployment on platforms with various resource constraints. Our approach comprises two stages. In the first stage, we enhance a pre-trained ViT with a nested elastic architecture that enables structural flexibility across MLP expansion ratio, number of attention heads, embedding dimension, and network depth. To preserve pre-trained knowledge and ensure stable adaptation, we adopt a curriculum-based training strategy that progressively increases elasticity. In the second stage, we design a lightweight router to select submodels according to computational budgets and downstream task demands. Initialized with Pareto-optimal configurations derived via a customized NSGA-II algorithm, the router is then jointly optimized with the backbone. Extensive experiments on multiple benchmarks demonstrate the effectiveness and versatility of EA-ViT. The code is available at https://github.com/zcxcf/EA-ViT.