Video large language models (LLMs) achieve strong video understanding by leveraging a large number of spatio-temporal tokens, but suffer from quadratic computational scaling with token count. To address this, we propose a training-free spatio-temporal token merging method, named STTM. Our key insight is to exploit local spatial and temporal redundancy in video data which has been overlooked in prior work. STTM first transforms each frame into multi-granular spatial tokens using a coarse-to-fine search over a quadtree structure, then performs directed pairwise merging across the temporal dimension. This decomposed merging approach outperforms existing token reduction methods across six video QA benchmarks. Notably, STTM achieves a 2x speed-up with only a 0.5% accuracy drop under a 50% token budget, and a 3x speed-up with just a 2% drop under a 30% budget. Moreover, STTM is query-agnostic, allowing KV cache reuse across different questions for the same video. The code is available at https://github.com/HYUNJS/STTM.
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
SpatialCrafter: Unleashing the Imagination of Video Diffusion Models for Scene Reconstruction from Limited Observations
PDF ↗Novel view synthesis (NVS) boosts immersive experiences in computer vision and graphics. Existing techniques, though progressed, rely on dense multi-view observations, restricting their application. This work takes on the challenge of reconstructing photorealistic 3D scenes from sparse or single-view inputs.We introduce SpatialCrafter, a framework that leverages the rich knowledge in video diffusion models to generate plausible additional observations, thereby alleviating reconstruction ambiguity. Through a trainable camera encoder and an epipolar attention mechanism for explicit geometric constraints, we achieve precise camera control and 3D consistency, further reinforced by a unified scale estimation strategy to handle scale discrepancies across datasets.Furthermore, by integrating monocular depth priors with semantic features in the video latent space, our framework directly regresses 3D Gaussian primitives and efficiently processes long-sequence features using a hybrid network structure. Extensive experiments show our method enhances sparse view reconstruction and restores the realistic appearance of 3D scenes.
Instruction-based video editing allows effective and interactive editing of videos using only instructions without extra inputs such as masks or attributes. However, collecting high-quality training triplets (source video, edited video, instruction) is a challenging task. Existing datasets mostly consist of low-resolution, short duration, and limited amount of source videos with unsatisfactory editing quality, limiting the performance of trained editing models. In this work, we present a high-quality Instruction-based Video Editing dataset with 1M triplets, namely InsViE-1M. We first curate high-resolution and high-quality source videos and images, then design an effective editing-filtering pipeline to construct high-quality editing triplets for model training. For a source video, we generate multiple edited samples of its first frame with different intensities of classifier-free guidance, which are automatically filtered by GPT-4o with carefully crafted guidelines. The edited first frame is propagated to subsequent frames to produce the edited video, followed by another round of filtering for frame quality and motion evaluation. We also generate and filter a variety of video editing triplets from high-quality images. With the InsViE-1M dataset, we propose a multi-stage learning strategy to train our InsViE model, progressively enhancing its instruction following and editing ability. Extensive experiments demonstrate the advantages of our InsViE-1M dataset and the trained model over state-of-the-art works. Data and code will be released.
Existing vector quantization (VQ) methods struggle with scalability, largely attributed to the instability of the codebook that undergoes partial updates during training. The codebook is prone to collapse as utilization decreases, due to the progressively widening distribution gap between non-activated codes and visual features. To solve the problem, we propose Index Backpropagation Quantization (IBQ), a new VQ method for the joint optimization of all codebook embeddings and the visual encoder. Applying a straight-through estimator on the one-hot categorical distribution between the encoded feature and codebook, all codes are differentiable and maintain a consistent latent space with the visual encoder. IBQ enables scalable training of visual tokenizers and, for the first time, achieves a large-scale codebook (2^ 18 ) with high dimension (256) and high utilization. Experiments on the standard ImageNet benchmark demonstrate the scalability and superiority of IBQ, achieving competitive results on reconstruction and the application of autoregressive visual generation.
Recent progress in multimodal large language models has markedly enhanced the understanding of short videos (typically under one minute), and several evaluation datasets have emerged accordingly. However, these advancements fall short of meeting the demands of real-world applications such as embodied intelligence for long-term decision-making, in-depth movie reviews and discussions, and live sports commentary, all of which require comprehension of long videos spanning several hours. To address this gap, we introduce LVBench, a benchmark specifically designed for long video understanding. Our dataset comprises publicly sourced videos and encompasses a diverse set of tasks aimed at long video comprehension and information extraction. LVBench is designed to challenge multimodal models to demonstrate long-term memory and extended comprehension capabilities. Our extensive evaluations reveal that current multimodal models still underperform on these demanding long video understanding tasks. Through LVBench, we aim to spur the development of more advanced models capable of tackling the complexities of long video comprehension.
Open-vocabulary object detection (OvOD) is set to revolutionize security screening by enabling systems to recognize any item in X-ray scans. However, developing effective OvOD models for X-ray imaging presents unique challenges due to data scarcity and the modality gap that prevents direct adoption of RGB-based solutions. To overcome these limitations, we propose RAXO, a training-free framework that repurposes off-the-shelf RGB OvOD detectors for robust X-ray detection. RAXO builds high-quality X-ray class descriptors using a dual-source retrieval strategy. It gathers relevant RGB images from the web and enriches them via a novel X-ray material transfer mechanism, eliminating the need for labeled databases. These visual descriptors replace text-based classification in OvOD, leveraging intra-modal feature distances for robust detection. Extensive experiments demonstrate that RAXO consistently improves OvOD performance, providing an average mAP increase of up to 17.0 points over base detectors. To further support research in this emerging field, we also introduce DET-COMPASS, a new benchmark featuring bounding box annotations for over 300 object categories, enabling large-scale evaluation of OvOD in X-ray. Code and dataset available at: https://pagf188.github.io/RAXO/.
Event cameras have shown promise in vision applications like optical flow estimation and stereo matching with many specialized architectures. However, existing works only focus event data within the confines of task-specific domains, overlooking the correlations between tasks across the temporal and spatial domains. In this paper, we propose a novel matching-based framework for event cameras to estimate flow and disparity simultaneously in a shared representation space, reformulating them as a unified pixel-wise correspondence matching problem. Specifically, our method utilizes a Temporal Recurrent Network to aggregate asynchronous event features across temporal or spatial domains, and a Spatial Contextual Attention to enhance knowledge transfer across event flows via temporal or spatial interactions. By utilizing a shared pixel-wise feature similarities module, our network performs optical flow estimation from temporal event segments and stereo matching from spatial event segments simultaneously. Our unified model inherently supports multi-task unification and cross-task transfer, which facilitate training and streamline deployment. Without the need for retraining on specific tasks, our model can effectively handle both event-based flow and stereo estimation, achieving state-of-the-art performance on both tasks. Our code will be released upon acceptance.
Vision foundation models (VFMs) such as DINOv2 and CLIP have achieved impressive results on various downstream tasks, but their limited feature resolution hampers performance in applications requiring pixel-level understanding. Feature upsampling offers a promising direction to address this challenge. In this work, we identify two critical factors for enhancing feature upsampling: the upsampler architecture and the training objective. For the upsampler architecture, we introduce a coordinate-based cross-attention transformer that integrates the high-resolution images with coordinates and low-resolution VFM features to generate sharp, high-quality features. For the training objective, we propose constructing high-resolution pseudo-groundtruth features by leveraging class-agnostic masks and self-distillation. Our approach effectively captures fine-grained details and adapts flexibly to various input and feature resolutions. Through experiments, we demonstrate that our approach significantly outperforms existing feature upsampling techniques across various downstream tasks. Our code is released at https://github.com/andrehuang/loftup.
An emerging paradigm in vision-and-language navigation (VLN) is the use of history-aware multi-modal transformer models. Given a language instruction, these models process observation and navigation history to predict the most appropriate action for an agent. While they have significantly improved performance, the scale of these models can be a bottleneck in practical settings with limited computational resources. In this work, we propose a novel input adaptive navigation method to enhance VLN model efficiency. We first show that existing input-adaptive mechanisms fail to reduce computations without substantial performance degradation. To address this, we introduce three adaptive algorithms, each deployed at a different level: (1) To improve spatial efficiency, we selectively process panoramic views at each observation of an agent. (2) To improve intra-model efficiency, we propose importance-based adaptive thresholding for the early-exit methods. (3) To improve temporal efficiency, we implement a caching mechanism that prevents reprocessing of views previously seen by the agent. In evaluations on seven VLN benchmarks, we demonstrate over a 2x reduction in computation across three off-the-shelf agents in both standard and continuous environments.
We introduce MOSCATO: a new benchmark for predicting the evolving states of multiple objects through long procedural videos with multiple actions. While prior work in object state prediction has typically focused on a single object undergoing one or a few state changes, real-world tasks require tracking many objects whose states evolve over multiple actions. Given the high cost of gathering framewise object-state labels for many videos, we develop a weakly-supervised multiple object state prediction framework, which only uses action labels during training. Specifically, we propose a novel Pseudo-Label Acquisition (PLA) pipeline that integrates large language models, vision-language models, and action segment annotations to generate fine-grained, per-frame object-state pseudo-labels for training a Multiple Object State Prediction (MOSP) network. We further devise a State-Action Interaction (SAI) module that explicitly models the correlations between actions and object states, thereby improving MOSP. To facilitate comprehensive evaluation, we create the MOSCATO benchmark by augmenting three egocentric video datasets with framewise object-state annotations. Experiments show that our multi-stage pseudo-labeling approach and SAI module significantly boost performance over zero-shot VLM baselines and naive extensions of existing methods, underscoring the importance of holistic action-state modeling for fine-grained procedural video understanding.
Data-Free Knowledge Distillation (DFKD) avoids accessing the original training data during knowledge transferring from a large model to a smaller one, possessing significant potential in ensuring the widespread promotion of industry-level applications while safeguarding user privacy and data security. Unfortunately, due to the lack of precise estimation of the original data distribution, existing DFKD methods often rely on manually induced priors to constrain the generator to produce samples that comply with the rules as much as possible. In this paper, we propose a novel method dubbed CouPling Network (CPNet) that constructs a generator to explicitly approximate the inverse transformation of the teacher model. Consequently, the two components can be integrated into an autoencoder specifically tailored for label information, where the generated images are treated as latent variables. Since real labels are typically uniformly distributed and the parameters of the teacher model are fixed, this enables our generator to produce images that closely approximate the true distribution. Besides, we transform real labels into feature-level constraints through the inverse transformation of a network classifier with fixed parameters, thereby converting the classification problem of generated images into an issue of distance measurement between features. We utilize this constraint for adversarial training and enhancing the diversity of produced images. Extensive experiments on three public benchmarks demonstrate that our proposed method achieves superior or competitive performance compared to previous state-of-the-art methods, while also exhibiting faster generation speed.
UniFuse: A Unified All-in-One Framework for Multi-Modal Medical Image Fusion Under Diverse Degradations and Misalignments
PDF ↗Current multimodal medical image fusion typically assumes that source images are of high quality and perfectly aligned at the pixel level. Its effectiveness heavily relies on these conditions and often deteriorates when handling misaligned or degraded medical images. To address this, we propose UniFuse, a general fusion framework. By embedding a degradation-aware prompt learning module, UniFuse seamlessly integrates multi-directional information from input images and correlates cross-modal alignment with restoration, enabling joint optimization of both tasks within a unified framework. Additionally, we design an Omni Unified Feature Representation scheme, which leverages Spatial Mamba to encode multi-directional features and mitigate modality differences in feature alignment. To enable simultaneous restoration and fusion within an All-in-One configuration, we propose a Universal Feature Restoration & Fusion module, incorporating the Adaptive LoRA Synergistic Network (ALSN) based on LoRA principles. By leveraging ALSN's adaptive feature representation along with degradation-type guidance, we enable joint restoration and fusion within a single-stage framework. Compared to staged approaches, UniFuse unifies alignment, restoration, and fusion within a single framework. Experimental results across multiple datasets demonstrate the method's effectiveness and significant advantages over existing approaches. code is available at https://github.com/slrl123/UniFuse.
As large-scale heterogeneous data sets become increasingly available, adapting foundation models at low cost has become a key issue. Seminal works in natural language processing, e.g. Low-Rank Adaptation (LoRA), leverage the low "intrinsic rank" of parameter updates during adaptation. In this paper, we argue that incorporating stronger inductive biases on both the data and the models can enhance the adaptation of Geospatial Foundation Models (GFMs), pretrained on RGB satellite images, to other types of optical satellite data. Specifically, the pretrained parameters of GFMs serve as a strong prior for the spatial structure of multispectral images. For this reason, we introduce DEFLECT (Deflecting Embeddings for Finetuning Latent representations for Earth and Climate Tasks), a novel strategy for adapting GFMs to multispectral satellite imagery with very few additional parameters. DEFLECT improves the representation capabilities of the extracted features, particularly enhancing spectral information, which is essential for geoscience and environmental-related tasks. We demonstrate the effectiveness of our method across three different GFMs and five diverse datasets, ranging from forest monitoring to marine environment segmentation. Compared to competing methods, DEFLECT achieves on-par or higher accuracy with 5-10x fewer parameters for classification and segmentation tasks. The code is available at https://github.com/VMarsocci/DEFLECT.
Modern machine learning models heavily rely on large datasets that often include sensitive and private information, raising serious privacy concerns. Differentially private (DP) data generation offers a solution by creating synthetic datasets that limit the leakage of private information within a predefined privacy budget; however, it requires a substantial amount of data to achieve performance comparable to models trained on the original data. To mitigate the significant expense incurred with synthetic data generation, Dataset Distillation (DD) stands out for its remarkable training and storage efficiency. This efficiency is particularly advantageous when integrated with DP mechanisms, curating compact yet informative synthetic datasets without compromising privacy. However, current state-of-the-art private DD methods suffer from a synchronized sampling-optimization process and the dependency on noisy training signals from randomly initialized networks. This results in the inefficient utilization of private information due to the addition of excessive noise. To address these issues, we introduce a novel framework that decouples sampling from optimization for better convergence and improves signal quality by mitigating the impact of DP noise through matching in an informative subspace. On CIFAR-10, our method achieves a 10.0% improvement with 50 images per class and 8.3% increase with just one-fifth the distilled set size of previous state-of-the-art methods, demonstrating significant potential to advance privacy-preserving DD.
The commercialization of generative artificial intelligence (GenAI) has led to a multi-level ecosystem involving model developers, service providers, and consumers. Thus, ensuring traceability is crucial, as service providers may violate intellectual property rights (IPR), and consumers may generate harmful content. However, existing methods are limited to single-level attribution scenarios and cannot simultaneously trace across multiple levels. To this end, we introduce a scalable dual fingerprinting method for text-to-image (T2I) models, to achieve traceability of both service providers and consumers. Specifically, we propose 2-headed Fingerprint-Informed Low-Rank Adaptation (FI-LoRA), where each head is controlled by a binary fingerprint and capable of introducing the fingerprints into generated images. In practice, one FI-LoRA head is used by the developer to assign a unique fingerprint to each service provider, while the other is made available to service providers for embedding consumer-specific fingerprints during image generation. Our method does not merely embed two fingerprints within the generated image but instead allows independent control over them at developer level and business level, enabling simultaneous traceability of businesses and consumers. Experiments show that our method applies to various image generation and editing tasks of multiple T2I models, and can achieve over 99.9% extraction accuracy for both fingerprints. Our method also demonstrates good robustness against both image-level attacks and white-box model-level attacks. We hope our work provides a unified solution for developers to implement multi-tiered traceability of their models and hierarchical control over model distribution and content generation.
AutoOcc: Automatic Open-Ended Semantic Occupancy Annotation via Vision-Language Guided Gaussian Splatting
PDF ↗Obtaining high-quality 3D semantic occupancy from raw sensor data remains an essential yet challenging task, often requiring extensive manual labeling. In this work, we propose AutoOcc, a vision-centric automated pipeline for open-ended semantic occupancy annotation that integrates differentiable Gaussian splatting guided by vision-language models. We formulate the open-ended semantic 3D occupancy reconstruction task to automatically generate scene occupancy by combining attention maps from vision-language models and foundation vision models. We devise semantic-aware Gaussians as intermediate geometric descriptors and propose a cumulative Gaussian-to-voxel splatting algorithm that enables effective and efficient occupancy annotation. Our framework outperforms existing automated occupancy annotation methods without human labels. AutoOcc also enables open-ended semantic occupancy auto-labeling, achieving robust performance in both static and dynamically complex scenarios.
Transferability makes the black-box attacks to be practical. Recent studies demonstrate that adversarial examples situated at the flat maxima on the loss landscape tend to exhibit higher transferability and propose effective strategies to optimize adversarial examples to converge toward that region. However, these works primarily consider the first-order gradient regularization and have yet to explore higher-order geometry properties of the flat loss landscape, which may lead to suboptimal results. In this work, we propose leveraging the trace of the Hessian matrix of loss function with respect to the adversarial example as a curvature-aware regularizer. For computationally efficient, we introduce an approximation method for the trace based on stochastic estimation and finite difference. We theoretically and empirically demonstrate that the trace of Hessian matrices for adversarial examples near local loss maxima is consistently negative. Following this insight, we propose **Negative Hessian Trace Regularization (NHTR)**, explicitly penalizing the negative Hessian trace to suppress curvature. Compared to existing first-order regularization methods, NHTR can generate adversarial examples at flatter local regions. Extensive experimental results on the ImageNet-compatible and CIFAR-10 datasets show that NHTR can significantly improve adversarial transferability than the state-of-the-art attacks.
Vision-language models (VLMs) often struggle with compositional reasoning due to insufficient high-quality image-text data. To tackle this challenge, we propose a novel block-based diffusion approach that automatically generates counterfactual datasets without manual annotation. Our method utilizes large language models to identify entities and their spatial relationships. It then independently generates image blocks as "puzzle pieces" coherently arranged according to specified compositional rules. This process creates diverse, high-fidelity counterfactual image-text pairs with precisely controlled variations. In addition, we introduce a specialized loss function that differentiates inter-set from intra-set samples, enhancing training efficiency and reducing the need for negative samples. Experiments demonstrate that fine-tuning VLMs with our counterfactual datasets significantly improves visual reasoning performance. Our approach achieves state-of-the-art results across multiple benchmarks while using substantially less training data than existing methods.
Panoramic image processing is essential for omni-context perception, yet faces constraints like distortions, perspective occlusions, and limited annotations. Previous unsupervised domain adaptation methods transfer knowledge from labeled pinhole data to unlabeled panoramic images, but they require access to source pinhole data. To address these, we introduce a more practical task, ie, Source-Free Occlusion-Aware Seamless Segmentation (SFOASS), and propose its first solution, called UNconstrained Learning Omni-Context Knowledge (UNLOCK). Specifically, UNLOCK includes two key modules: Omni Pseudo-Labeling Learning and Amodal-Driven Context Learning. While adapting without relying on source data or target labels, this framework enhances models to achieve segmentation with 360deg viewpoint coverage and occlusion-aware reasoning. Furthermore, we benchmark the proposed SFOASS task through both real-to-real and synthetic-to-real adaptation settings. Experimental results show that our source-free method achieves performance comparable to source-dependent methods, yielding SOTA scores of 10.9 in mAAP and 11.6 in mAP, along with an absolute improvement of +4.3 in mAPQ over the source-only method. All data and code will be made publicly available at https://github.com/yihong-97/UNLOCK.
Large pre-trained transformers have revolutionized artificial intelligence across various domains, and fine-tuning remains the dominant approach for adapting these models to downstream tasks due to the cost of training from scratch. However, in existing fine-tuning methods, the updated representations are formed as a dense combination of modified parameters, making it challenging to interpret their contributions and understand how the model adapts to new tasks.In this work, we introduce a fine-tuning framework inspired by sparse coding, where fine-tuned features are represented as a sparse combination of basic elements, i.e., feature dictionary atoms. Sparse coefficients then serve as indicators of atom importance, identifying the contribution of each atom to the updated representation.The feature dictionary atoms function as fundamental building blocks of the representation, and tuning atoms allows for seamless adaptation to downstream tasks.Leveraging the atom selection capability of sparse coefficients, we first demonstrate that our method enhances image editing performance by improving text alignment through the removal of unimportant feature dictionary atoms.Additionally, we validate the effectiveness of our approach in the text-to-image concept customization task, where our method efficiently constructs the target concept using a sparse combination of feature dictionary atoms, outperforming various baseline fine-tuning methods.