3D point clouds are increasingly vital for applications like autonomous driving and robotics, yet the raw data captured by sensors often suffer from noise and sparsity, creating challenges for downstream tasks. Consequently, point cloud upsampling becomes essential for improving density and uniformity, with recent approaches showing promise by projecting randomly generated query points onto the underlying surface of sparse point clouds. However, these methods often result in outliers, non-uniformity, and difficulties in handling regions with high curvature and intricate structures. In this work, we address these challenges by introducing the Progressive Local Surface Estimator (PLSE), which more effectively captures local features in complex regions through a curvature-based sampling technique that selectively targets high-curvature areas. Additionally, we incorporate a curriculum learning strategy that leverages the curvature distribution within the point cloud to naturally assess the sample difficulty, enabling curriculum learning on point cloud data for the first time. The experimental results demonstrate that our approach significantly outperforms existing methods, achieving high-quality, dense point clouds with superior accuracy and detail.

Open-Vocabulary Semantic Segmentation (OVSS) has advanced with recent vision-language models (VLMs), enabling segmentation beyond predefined categories through various learning schemes. Notably, training-free methods offer scalable, easily deployable solutions for handling unseen data, a key goal of OVSS. Yet, a critical issue persists: lack of object-level context consideration when segmenting complex objects in the challenging environment of OVSS based on arbitrary query prompts. This oversight limits models’ ability to group semantically consistent elements within object and map them precisely to user-defined arbitrary classes. In this work, we introduce a novel approach that overcomes this limitation by incorporating object-level contextual knowledge within images. Specifically, our model enhances intra-object consistency by distilling spectral-driven features from vision foundation models into the attention mechanism of the visual encoder, enabling semantically coherent components to form a single object mask. Additionally, we refine the text embeddings with zero-shot object presence likelihood to ensure accurate alignment with the specific objects represented in the images. By leveraging object-level contextual knowledge, our proposed approach achieves state-of-the-art performance with strong generalizability across diverse datasets.

Diffusion-based text-to-image (T2I) models have recently excelled in high-quality image generation, particularly in a training-free manner, enabling cost-effective adaptability and generalization across diverse tasks. However, while the existing methods have been continuously focusing on several challenges such as "missing objects" and "mismatched attributes," another critical issue of "mislocated objects" remains where generated spatial positions fail to align with text prompts. Surprisingly, ensuring such seemingly basic functionality remains challenging in popular T2I models due to the inherent difficulty of imposing explicit spatial guidance via text forms. To address this, we propose STORM (Spatial Transport Optimization by Repositioning Attention Map), a novel training-free approach for spatially coherent T2I synthesis. STORM employs Spatial Transport Optimization (STO), rooted in optimal transport theory, to dynamically adjust object attention maps for precise spatial adherence, supported by a Spatial Transport (ST) Cost function that enhances spatial understanding. Our analysis shows that integrating spatial awareness is most effective in the early denoising stages, while later phases refine details. Extensive experiments demonstrate that STORM surpasses existing methods, effectively mitigating mislocated objects while improving missing and mismatched attributes, setting a new benchmark for spatial alignment in T2I synthesis.

Visual grounding seeks to localize the image region corresponding to a free-form text description. Recently, the strong multimodal capabilities of Large Vision-Language Models (LVLMs) have driven substantial improvements in visual grounding, though they inevitably require fine-tuning and additional model components to explicitly generate bounding boxes or segmentation masks. However, we discover that a few attention heads in frozen LVLMs demonstrate strong visual grounding capabilities. We refer to these heads, which consistently capture object locations related to text semantics, as localization heads. Using localization heads, we introduce a straightforward and effective training-free visual grounding framework that utilizes text-to-image attention maps from localization heads to identify the target objects. Surprisingly, only three out of thousands of attention heads are sufficient to achieve competitive localization performance compared to existing LVLM-based visual grounding methods that require fine-tuning. Our findings suggest that LVLMs can innately ground objects based on a deep comprehension of the text-image relationship, as they implicitly focus on relevant image regions to generate informative text outputs.

Generating high-quality novel views of a scene from a single image requires maintaining structural coherence across different views, referred to as view consistency. While diffusion models have driven advancements in novel view synthesis, they still struggle to preserve spatial continuity across views. Diffusion models have been combined with 3D models to address the issue, but such approaches lack efficiency due to their complex multi-step pipelines. This paper proposes a novel view-consistent image generation method which utilizes diffusion models without additional modules. Our key idea is to enhance diffusion models with a training-free method that enables adaptive attention manipulation and noise reinitialization by leveraging view-guided warping to ensure view consistency. Through our comprehensive metric framework suitable for novel-view datasets, we show that our method improves view consistency across various diffusion models, demonstrating its broader applicability.

Large multimodal models (LMMs) "see" images by leveraging the attention mechanism between text and visual tokens in the transformer decoder. Ideally, these models should focus on key visual information relevant to the text token. However, recent findings indicate that LMMs have an extraordinary tendency to consistently allocate high attention weights to specific visual tokens, even when these tokens are irrelevant to the corresponding text. In this study, we investigate the property behind the appearance of these irrelevant visual tokens and examine their characteristics. Our findings show that this behavior arises due to the massive activation of certain hidden state dimensions, which resembles the attention sink found in language models. Hence, we refer to this phenomenon as the visual attention sink. In particular, our analysis reveals that removing the irrelevant visual sink tokens does not impact model performance, despite receiving high attention weights. Consequently, we recycle the attention to these tokens as surplus resources, redistributing the attention budget to enhance focus on the image. To achieve this, we introduce Visual Attention Redistribution (VAR), a method that redistributes attention in image-centric heads, which we identify as innately focusing on visual information. VAR can be seamlessly applied across different LMMs to improve performance on a wide range of tasks, including general vision-language tasks, visual hallucination tasks, and vision-centric tasks, all without the need for additional training, models, or inference steps. Experimental results demonstrate that VAR enables LMMs to process visual information more effectively by adjusting their internal attention mechanisms, offering a new direction to enhancing the multimodal capabilities of LMMs.

Starting from flow- and diffusion-based transformers, Multi-modal Diffusion Transformers (MM-DiTs) have reshaped text-to-vision generation, gaining acclaim for exceptional visual fidelity. As these models advance, users continually push the boundary with imaginative or rare prompts, which advanced models still falter in generating, since their concepts are often too scarce to leave a strong imprint during pre-training. In this paper, we propose a simple yet effective intervention that surfaces rare semantics inside MM-DiTs without additional training steps, data, denoising-time optimization, or reliance on external modules (e.g., large language models). In particular, the joint-attention mechanism intrinsic to MM-DiT sequentially updates text embeddings alongside image embeddings throughout transformer blocks. We find that by mathematically expanding representational basins around text token embeddings via variance scale-up before the joint-attention blocks, rare semantics clearly emerge in MM-DiT’s outputs. Furthermore, our results generalize effectively across text-to-vision tasks, including text-to-image, text-to-video, and text-driven image editing. Our work invites generative models to reveal the semantics that users intend, once hidden yet ready to surface.

How do vision transformers (ViTs) represent and process the world? This paper addresses this long-standing question through the first systematic analysis of 6.6K features across all layers, extracted via sparse autoencoders, and by introducing the residual replacement model, which replaces ViT computations with interpretable features in the residual stream. Our analysis reveals not only a feature evolution from low-level patterns to high-level semantics, but also how ViTs encode curves and spatial positions through specialized feature types. The residual replacement model scalably produces a faithful yet parsimonious circuit for human-scale interpretability by significantly simplifying the original computations. As a result, this framework enables intuitive understanding of ViT mechanisms. Finally, we demonstrate the utility of our framework in debiasing spurious correlations.

Point-based image editing enables accurate and flexible control through content dragging. However, the role of text embedding during the editing process has not been thoroughly investigated. A significant aspect that remains unexplored is the interaction between text and image embeddings. During the progressive editing in a diffusion model, the text embedding remains constant. As the image embedding increasingly diverges from its initial state, the discrepancy between the image and text embeddings presents a significant challenge. In this study, we found that the text prompt significantly influences the dragging process, particularly in maintaining content integrity and achieving the desired manipulation. Upon these insights, we propose Dragtext,which optimizes text embedding in conjunction with the dragging process to pair with the modified image embedding. Simultaneously, we regularize the text optimization process to preserve the integrity of the original text prompt. Our approach can be seamlessly integrated with existing diffusion-based drag methods, enhancing performance with only a few lines of code.

Disentangled representation learning (DRL) aims to break down observed data into core intrinsic factors for a profound understanding of the data. In real-world scenarios, manually defining and labeling these factors are non-trivial, making unsupervised methods attractive. Recently, there have been limited explorations of utilizing diffusion models (DMs), which are already mainstream in generative modeling, for unsupervised DRL. They implement their own inductive bias to ensure that each latent unit input to the DM expresses only one distinct factor. In this context, we design Dynamic Gaussian Anchoring to enforce attribute-separated latent units for more interpretable DRL. This unconventional inductive bias explicitly delineates the decision boundaries between attributes while also promoting the independence among latent units. Additionally, we also propose Skip Dropout technique, which easily modifies the denoising U-Net to be more DRL-friendly, addressing its uncooperative nature with the disentangling feature extractor. Our methods, which carefully consider the latent unit semantics and the distinct DM structure, enhance the practicality of DM-based disentangled representations, demonstrating state-of-the-art disentanglement performance on both synthetic and real data, as well as advantages in downstream tasks.

Semantic segmentation has innately relied on extensive pixel-level annotated data, leading to the emergence of unsupervised methodologies. Among them, leveraging self-supervised Vision Transformers for unsupervised semantic segmentation (USS) has been making steady progress with expressive deep features. Yet, for semantically segmenting images with complex objects, a predominant challenge remains: the lack of explicit object-level semantic encoding in patch-level features. This technical limitation often leads to inadequate segmentation of complex objects with diverse structures. To address this gap, we present a novel approach, EAGLE, which emphasizes object-centric representation learning for unsupervised semantic segmentation. Specifically, we introduce EiCue, a spectral technique providing semantic and structural cues through an eigenbasis derived from the semantic similarity matrix of deep image features and color affinity from an image. Further, by incor-porating our object-centric contrastive loss with EiCue, we guide our model to learn object-level representations with intra- and inter-image object-feature consistency, thereby enhancing semantic accuracy. Extensive experiments on COCO-Stuff, Cityscapes, and Potsdam-3 datasets demonstrate the state-of-the-art USS results of EAGLE with accurate and consistent semantic segmentation across complex scenes.

Large-scale data obtained from aggregation of already collected multi-site neuroimaging datasets has brought benefits such as higher statistical power, reliability, and robustness to the studies. Despite these promises from growth in sample size, substantial technical variability stemming from differences in scanner specifications exists in the aggregated data and could inadvertently bias any downstream analyses on it. Such a challenge calls for data normalization and/or harmonization frameworks, in addition to comprehensive criteria to estimate the scanner-related variability and evaluate the harmonization frameworks. In this study, we propose MISPEL (Multi-scanner Image harmonization via Structure Preserving Embedding Learning), a supervised multi-scanner harmonization method that is naturally extendable to more than two scanners. We also designed a set of criteria to investigate the scanner-related technical variability and evaluate the harmonization techniques. As an essential requirement of our criteria, we introduced a multi-scanner matched dataset of 3T T1 images across four scanners, which, to the best of our knowledge is one of the few datasets of this kind. We also investigated our evaluations using two popular segmentation frameworks: FSL and segmentation in statistical parametric mapping (SPM). Lastly, we compared MISPEL to popular methods of normalization and harmonization, namely White Stripe, RAVEL, and CALAMITI. MISPEL outperformed these methods and is promising for many other neuroimaging modalities.

Abstract Theoretically, domain adaptation is a well-researched problem. Further, this theory has been well-used in practice. In particular, we note the bound on target error given by Ben-David et al. (Mach Learn 79(1–2):151–175, 2010) and the well-known domain-aligning algorithm based on this work using Domain Adversarial Neural Networks (DANN) presented by Ganin and Lempitsky (in International conference on machine learning, pp 1180–1189). Recently, multiple variants of DANN have been proposed for the related problem of domain generalization , but without much discussion of the original motivating bound. In this paper, we investigate the validity of DANN in domain generalization from this perspective. We investigate conditions under which application of DANN makes sense and further consider DANN as a dynamic process during training. Our investigation suggests that the application of DANN to domain generalization may not be as straightforward as it seems. To address this, we design an algorithmic extension to DANN in the domain generalization case. Our experimentation validates both theory and algorithm.

The topic of generalizing machine learning models learned on a collection of source domains to unknown target domains is challenging. While many domain generalization (DG) methods have achieved promising results, they primarily rely on the source domains at train-time without manipulating the target domains at test-time. Thus, it is still possible that those methods can overfit to source domains and perform poorly on target domains. Driven by the observation that domains are strongly related to styles, we argue that reducing the gap between source and target styles can boost models’ generalizability. To solve the dilemma of having no access to the target domain during training, we introduce Test-time Fourier Style Calibration (TF-Cal) for calibrating the target domain style on the fly during testing. To access styles, we utilize Fourier transformation to decompose features into amplitude (style) features and phase (semantic) features. Furthermore, we present an effective technique to Augment Amplitude Features (AAF) to complement TF-Cal. Extensive experiments on several popular DG benchmarks and a segmentation dataset for medical images demonstrate that our method outperforms state-of-the-art methods.

Multiclass neural networks are a common tool in modern unsupervised domain adaptation, yet an appropriate theoretical description for their non-uniform sample complexity is lacking in the adaptation literature. To fill this gap, we propose the first PAC-Bayesian adaptation bounds for multiclass learners. We facilitate practical use of our bounds by also proposing the first approximation techniques for the multiclass distribution divergences we consider. For divergences dependent on a Gibbs predictor, we propose additional PAC-Bayesian adaptation bounds which remove the need for inefficient Monte-Carlo estimation. Empirically, we test the efficacy of our proposed approximation techniques as well as some novel design-concepts which we include in our bounds. Finally, we apply our bounds to analyze a common adaptation algorithm that uses neural networks.

Despite the extensive usage of point clouds in 3D vision, relatively limited data are available for training deep neural networks. Although data augmentation is a standard approach to compensate for the scarcity of data, it has been less explored in the point cloud literature. In this paper, we propose a simple and effective augmentation method called PointWOLF for point cloud augmentation. The proposed method produces smoothly varying non-rigid deformations by locally weighted transformations centered at multiple anchor points. The smooth deformations allow diverse and realistic augmentations. Furthermore, in order to minimize the manual efforts to search the optimal hyperparameters for augmentation, we present AugTune, which generates augmented samples of desired difficulties producing targeted confidence scores. Our experiments show our framework consistently improves the performance for both shape classification and part segmentation tasks. Particularly, with PointNet++, PointWOLF achieves the state-of-the-art 89.7 accuracy on shape classification with the real-world ScanObjectNN dataset. The code is available at https://github.com/mlvlab/PointWOLF.

We develop a conditional generative model for longitudinal image datasets based on sequential invertible neural networks. Longitudinal image acquisitions are common in various scientific and biomedical studies where often each image sequence sample may also come together with various secondary (fixed or temporally dependent) measurements. The key goal is not only to estimate the parameters of a deep generative model for the given longitudinal data, but also to enable evaluation of how the temporal course of the generated longitudinal samples are influenced as a function of induced changes in the (secondary) temporal measurements (or events). Our proposed formulation incorporates recurrent subnetworks and temporal context gating, which provide a smooth transition in a temporal sequence of generated data that can be easily informed or modulated by secondary temporal conditioning variables. We show that the formulation works well despite the smaller sample sizes common in these applications. Our model is validated on two video datasets and a longitudinal Alzheimer's disease (AD) dataset for both quantitative and qualitative evaluations of the generated samples. Further, using our generated longitudinal image samples, we show that we can capture the pathological progressions in the brain that turn out to be consistent with the existing literature, and could facilitate various types of downstream statistical analysis.

Video is one of the richest sources of information available online but extracting deep insights from video content at internet scale is still an open problem, both in terms of depth and breadth of understanding, as well as scale. Over the last few years, the field of video understanding has made great strides due to the availability of large-scale video datasets and core advances in image, audio, and video modeling architectures. However, the state-of-the-art architectures on small scale datasets are frequently impractical to deploy at internet scale, both in terms of the ability to train such deep networks on hundreds of millions of videos, and to deploy them for inference on billions of videos. In this paper, we present a MapReduce-based training framework, which exploits both data parallelism and model parallelism to scale training of complex video models. The proposed framework uses alternating optimization and full-batch fine-tuning, and supports large Mixture-of-Experts classifiers with hundreds of thousands of mixtures, which enables a trade-off between model depth and breadth, and the ability to shift model capacity between shared (generalization) layers and per-class (specialization) layers. We demonstrate that the proposed framework is able to reach state-of-the-art performance on the largest public video datasets, YouTube-8M and Sports-1M, and can scale to 100 times larger datasets.

Visual relationships provide higher-level information of objects and their relations in an image - this enables a semantic understanding of the scene and helps downstream applications. Given a set of localized objects in some training data, visual relationship detection seeks to detect the most likely "relationship" between objects in a given image. While the specific objects may be well represented in training data, their relationships may still be infrequent. The empirical distribution obtained from seeing these relationships in a dataset does not model the underlying distribution well - a serious issue for most learning methods. In this work, we start from a simple multi-relational learning model, which in principle, offers a rich formalization for deriving a strong prior for learning visual relationships. While the inference problem for deriving the regularizer is challenging, our main technical contribution is to show how adapting recent results in numerical linear algebra lead to efficient algorithms for a factorization scheme that yields highly informative priors. The factorization provides sample size bounds for inference (under mild conditions) for the underlying [object, predicate, object] relationship learning task on its own and surprisingly outperforms (in some cases) existing methods even without utilizing visual features. Then, when integrated with an end-to-end architecture for visual relationship detection leveraging image data, we substantially improve the state-of-the-art.

The adoption of "human-in-the-loop" paradigms in computer vision and machine learning is leading to various applications where the actual data acquisition (e.g., human supervision) and the underlying inference algorithms are closely interwined. While classical work in active learning provides effective solutions when the learning module involves classification and regression tasks, many practical issues such as partially observed measurements, financial constraints and even additional distributional or structural aspects of the data typically fall outside the scope of this treatment. For instance, with sequential acquisition of partial measurements of data that manifest as a matrix (or tensor), novel strategies for completion (or collaborative filtering) of the remaining entries have only been studied recently. Motivated by vision problems where we seek to annotate a large dataset of images via a crowdsourced platform or alternatively, complement results from a state-of-the-art object detector using human feedback, we study the "completion" problem defined on graphs, where requests for additional measurements must be made sequentially. We design the optimization model in the Fourier domain of the graph describing how ideas based on adaptive submodularity provide algorithms that work well in practice. On a large set of images collected from Imgur, we see promising results on images that are otherwise difficult to categorize. We also show applications to an experimental design problem in neuroimaging.

There is a great deal of interest in using large scale brain imaging studies to understand how brain connectivity evolves over time for an individual and how it varies over different levels/quantiles of cognitive function. To do so, one typically performs so-called tractography procedures on diffusion MR brain images and derives measures of brain connectivity expressed as graphs. The nodes correspond to distinct brain regions and the edges encode the strength of the connection. The scientific interest is in characterizing the evolution of these graphs over time or from healthy individuals to diseased. We pose this important question in terms of the Laplacian of the connectivity graphs derived from various longitudinal or disease time points - quantifying its progression is then expressed in terms of coupling the harmonic bases of a full set of Laplacians. We derive a coupled system of generalized eigenvalue problems (and corresponding numerical optimization schemes) whose solution helps characterize the full life cycle of brain connectivity evolution in a given dataset. Finally, we show a set of results on a diffusion MR imaging dataset of middle aged people at risk for Alzheimer's disease (AD), who are cognitively healthy. In such asymptomatic adults, we find that a framework for characterizing brain connectivity evolution provides the ability to predict cognitive scores for individual subjects, and for estimating the progression of participant's brain connectivity into the future.

Eigenvalue problems are ubiquitous in computer vision, covering a very broad spectrum of applications ranging from estimation problems in multi-view geometry to image segmentation. Few other linear algebra problems have a more mature set of numerical routines available and many computer vision libraries leverage such tools extensively. However, the ability to call the underlying solver only as a "black box" can often become restrictive. Many 'human in the loop' settings in vision frequently exploit supervision from an expert, to the extent that the user can be considered a subroutine in the overall system. In other cases, there is additional domain knowledge, side or even partial information that one may want to incorporate within the formulation. In general, regularizing a (generalized) eigenvalue problem with such side information remains difficult. Motivated by these needs, this paper presents an optimization scheme to solve generalized eigenvalue problems (GEP) involving a (nonsmooth) regularizer. We start from an alternative formulation of GEP where the feasibility set of the model involves the Stiefel manifold. The core of this paper presents an end to end stochastic optimization scheme for the resultant problem. We show how this general algorithm enables improved statistical analysis of brain imaging data where the regularizer is derived from other 'views' of the disease pathology, involving clinical measurements and other image-derived representations.