Structure-from-Motion (SfM) is a fundamental 3D vision task for recovering camera parameters and scene geometry from multi-view images. While recent deep learning advances enable accurate Monocular Depth Estimation (MDE) from single images without depending on camera motion, integrating MDE into SfM remains a challenge. Unlike conventional triangulated sparse point clouds, MDE produces dense depth maps with significantly higher error variance. Inspired by modern RANSAC estimators, we propose Marginalized Bundle Adjustment (MBA) to mitigate MDE error variance leveraging its density. With MBA, we show that MDE depth maps are sufficiently accurate to yield SoTA or competitive results in SfM and camera relocalization tasks. Through extensive evaluations, we demonstrate consistently robust performance across varying scales, ranging from few-frame setups to large multiview systems with thousands of images. Our method highlights the significant potential of MDE in multi-view 3D vision. Code is available at https://marginalizedba.github.io/.

Long video understanding presents challenges due to the inherent high computational complexity and redundant temporal information. An effective representation for long videos must efficiently process such redundancy while preserving essential contents for downstream tasks. This paper introduces SEmantic Attention Learning (SEAL), a novel unified representation for long videos. To reduce computational complexity, long videos are decomposed into three distinct types of semantic entities: scenes, objects, and actions, allowing models to operate on a compact set of entities rather than a large number of frames or pixels. To further address redundancy, we propose an attention learning module that balances token relevance with diversity, formulated as a subset selection optimization problem. Our representation is versatile and applicable across various long video understanding tasks. Extensive experiments demonstrate that SEAL significantly outperforms state-of-the-art methods in video question answering and temporal grounding tasks across diverse benchmarks, including LVBench, MovieChat-1K, and Ego4D.

Generative models transform random noise into images, while their inversion aims to reconstruct structured noise for recovery and editing. This paper addresses two key tasks: (i) *inversion* and (ii) *editing* of real images using stochastic equivalents of rectified flow models (e.g., Flux). While Diffusion Models (DMs) dominate the field of generative modeling for images, their inversion suffers from faithfulness and editability challenges due to nonlinear drift and diffusion. Existing DM inversion methods require costly training of additional parameters or test-time optimization of latent variables. Rectified Flows (RFs) offer a promising alternative to DMs, yet their inversion remains underexplored. We propose RF inversion using dynamic optimal control derived via a linear quadratic regulator, and prove that the resulting vector field is equivalent to a rectified stochastic differential equation. We further extend our framework to design a stochastic sampler for Flux. Our method achieves state-of-the-art performance in zero-shot inversion and editing, surpassing prior works in stroke-to-image synthesis and semantic image editing, with large-scale human evaluations confirming user preference. See our project page https://rf-inversion.github.io/ for code and demo.

We propose Reference-Based Modulation (RB-Modulation), a new plug-and-play solution for training-free personalization of diffusion models. Existing training-free approaches exhibit difficulties in (a) style extraction from reference images in the absence of additional style or content text descriptions, (b) unwanted content leakage from reference style images, and (c) effective composition of style and content. RB-Modulation is built on a novel stochastic optimal controller where a style descriptor encodes the desired attributes through a terminal cost. The resulting drift not only overcomes the difficulties above, but also ensures high fidelity to the reference style and adheres to the given text prompt. We also introduce a cross-attention-based feature aggregation scheme that allows RB-Modulation to decouple content and style from the reference image. With theoretical justification and empirical evidence, our test-time optimization framework demonstrates precise extraction and control of *content* and *style* in a training-free manner. Further, our method allows a seamless composition of content and style, which marks a departure from the dependency on external adapters or ControlNets. See project page: https://rb-modulation.github.io/ for code and further details.

Sampling from the posterior distribution in latent diffusion models for inverse problems is computationally challenging. Existing methods often rely on Tweedie's first-order moments that tend to induce biased results [32]. Second-order approximations are computationally prohibitive, making standard reverse diffusion processes in-tractable for posterior sampling. We present Second-order Tweedie sampler from Surrogate Loss (STSL), a novel sampler offering efficiency comparable to first-order Tweedie while enabling tractable reverse processes using second-order approximation. Theoretical results reveal that our approach establishes a lower bound through a surrogate loss and enables a tractable reverse process using the trace of the Hessian with only$\mathcal{O}(1)$compute. We show STSL out-performs SoTA solvers PSLD [43] and P2L [10] by reducing neural function evaluations by 4X and 8X, respectively, while enhancing sampling quality on FFHQ, ImageNet, and COCO benchmarks. Moreover, STSL extends to text-guided image editing, effectively mitigating residual distortions in corrupted images. To our best knowledge, this is the first work to offer an efficient second-order approximation for solving inverse problems using latent diffusion, which further enables editing real-world images with corruptions.

This work studies the generalization issue of face anti-spoofing (FAS) models on domain gaps, such as image resolution, blurriness and sensor variations. Most prior works regard domain-specific signals as a negative impact, and apply metric learning or adversarial losses to remove them from feature representation. Though learning a domain-invariant feature space is viable for the training data, we show that the feature shift still exists in an unseen test domain, which backfires on the generalizability of the classifier. In this work, instead of constructing a domain-invariant feature space, we encourage domain separability while aligning the live-to-spoof transition (i.e., the trajectory from live to spoof) to be the same for all domains. We formulate this FAS strategy of separability and alignment (SA-FAS) as a problem of invariant risk minimization (IRM), and learn domain-variant feature representation but domain-invariant classifier. We demonstrate the effectiveness of SA-FAS on challenging cross-domain FAS datasets and establish state-of-the-art performance. Code is available at https://github.com/sunyiyou/SAFAS.

Presentation attack detection (PAD) is a critical component in secure face authentication. We present a PAD algorithm to distinguish face spoofs generated by a photograph of a subject from live images. Our method uses an image decomposition network to extract albedo and normal. The domain gap between the real and spoof face images leads to easily identifiable differences, especially between the re-covered albedo maps. We enhance this domain gap by retraining existing methods using supervised contrastive loss. We present empirical and theoretical analysis that demonstrates that contrast and lighting effects can play a significant role in PAD; these show up particularly in the recovered albedo. Finally, we demonstrate that by combining all of these methods we achieve state-of-the-art results on both intra-dataset testing for CelebA-Spoof, OULU, CASIA-SURF datasets and inter-dataset setting on SiW, CASIA-MFSD, Replay-Attack and MSU-MFSD datasets.

We present Retrieve in Style (RIS), an unsupervised framework for facial feature transfer and retrieval on real images. Recent work shows capabilities of transferring local facial features by capitalizing on the disentanglement property of the StyleGAN latent space. RIS improves existing art on the following: 1) Introducing more effective feature disentanglement to allow for challenging transfers (i.e., hair, pose) that were not shown possible in SoTA methods. 2) Eliminating the need for per-image hyperparameter tuning, and for computing a catalog over a large batch of images. 3) Enabling fine-grained face retrieval using disentangled facial features (e.g., eyes). To our best knowledge, this is the first work to retrieve face images at this fine level. 4) Demonstrating robust, natural editing on real images. Our qualitative and quantitative analyses show RIS achieves both high-fidelity feature transfers and accurate fine-grained retrievals on real images. We also discuss the responsible applications of RIS. Our code is available at https://github.com/mchong6/RetrieveInStyle.

We present a scalable weakly supervised clustering approach to learn facial action units (AUs) from large, freely available web images. Unlike most existing methods (e.g., CNNs) that rely on fully annotated data, our method exploits web images with inaccurate annotations. Specifically, we derive a weakly-supervised spectral algorithm that learns an embedding space to couple image appearance and semantics. The algorithm has efficient gradient update, and scales up to large quantities of images with a stochastic extension. With the learned embedding space, we adopt rank-order clustering to identify groups of visually and semantically similar images, and re-annotate these groups for training AU classifiers. Evaluation on the 1 millon EmotioNet dataset demonstrates the effectiveness of our approach: (1) our learned annotations reach on average 91.3% agreement with human annotations on 7 common AUs, (2) classifiers trained with re-annotated images perform comparably to, sometimes even better than, its supervised CNN-based counterpart, and (3) our method offers intuitive outlier/noise pruning instead of forcing one annotation to every image. Code is available.

Action unit detection in infants relative to adults presents unique challenges. Jaw contour is less distinct, facial texture is reduced, and rapid and unusual facial movements are common. To detect facial action units in spontaneous behavior of infants, we propose a multi-label Convolutional Neural Network (CNN). Eighty-six infants were recorded during tasks intended to elicit enjoyment and frustration. Using an extension of FACS for infants (Baby FACS), over 230,000 frames were manually coded for ground truth. To control for chance agreement, inter-observer agreement between Baby-FACS coders was quantified using free-margin kappa. Kappa coefficients ranged from 0.79 to 0.93, which represents high agreement. The multi-label CNN achieved comparable agreement with manual coding. Kappa ranged from 0.69 to 0.93. Importantly, the CNN-based AU detection revealed the same change in findings with respect to infant expressiveness between tasks. While further research is needed, these findings suggest that automatic AU detection in infants is a viable alternative to manual coding of infant facial expression.

Facial action units (AU) are the fundamental units to decode human facial expressions. At least three aspects affect performance of automated AU detection: spatial representation, temporal modeling, and AU correlation. Unlike most studies that tackle these aspects separately, we propose a hybrid network architecture to jointly model them. Specifically, spatial representations are extracted by a Convolutional Neural Network (CNN), which, as analyzed in this paper, is able to reduce person-specific biases caused by hand-crafted descriptors (e.g., HOG and Gabor). To model temporal dependencies, Long Short-Term Memory (LSTMs) are stacked on top of these representations, regardless of the lengths of input videos. The outputs of CNNs and LSTMs are further aggregated into a fusion network to produce per-frame prediction of 12 AUs. Our network naturally addresses the three issues together, and yields superior performance compared to existing methods that consider these issues independently. Extensive experiments were conducted on two large spontaneous datasets, GFT and BP4D, with more than 400,000 frames coded with 12 AUs. On both datasets, we report improvements over a standard multi-label CNN and feature-based state-of-the-art. Finally, we provide visualization of the learned AU models, which, to our best knowledge, reveal how machines see AUs for the first time.

Despite the important role that facial expressions play in interpersonal communication and our knowledge that interpersonal behavior is influenced by social context, no currently available facial expression database includes multiple interacting participants. The Sayette Group Formation Task (GFT) database addresses the need for well-annotated video of multiple participants during unscripted interactions. The database includes 172,800 video frames from 96 participants in 32 three-person groups. To aid in the development of automated facial expression analysis systems, GFT includes expert annotations of FACS occurrence and intensity, facial landmark tracking, and baseline results for linear SVM, deep learning, active patch learning, and personalized classification. Baseline performance is quantified and compared using identical partitioning and a variety of metrics (including means and confidence intervals). The highest performance scores were found for the deep learning and active patch learning methods. Learn more at http://osf.io/7wcyz.

Automatic facial action unit (AU) and expression detection from videos is a long-standing problem. The problem is challenging in part because classifiers must generalize to previously unknown subjects that differ markedly in behavior and facial morphology (e.g., heavy versus delicate brows, smooth versus deeply etched wrinkles) from those on which the classifiers are trained. While some progress has been achieved through improvements in choices of features and classifiers, the challenge occasioned by individual differences among people remains. Person-specific classifiers would be a possible solution but for a paucity of training data. Sufficient training data for person-specific classifiers typically is unavailable. This paper addresses the problem of how to personalize a generic classifier without additional labels from the test subject. We propose a transductive learning method, which we refer to as a Selective Transfer Machine (STM), to personalize a generic classifier by attenuating person-specific mismatches. STM achieves this effect by simultaneously learning a classifier and re-weighting the training samples that are most relevant to the test subject. We compared STM to both generic classifiers and cross-domain learning methods on four benchmarks: CK+ [44], GEMEP-FERA [67], RUFACS [4] and GFT [57]. STM outperformed generic classifiers in all.

Region learning (RL) and multi-label learning (ML) have recently attracted increasing attentions in the field of facial Action Unit (AU) detection. Knowing that AUs are active on sparse facial regions, RL aims to identify these regions for a better specificity. On the other hand, a strong statistical evidence of AU correlations suggests that ML is a natural way to model the detection task. In this paper, we propose Deep Region and Multi-label Learning (DRML), a unified deep network that simultaneously addresses these two problems. One crucial aspect in DRML is a novel region layer that uses feed-forward functions to induce important facial regions, forcing the learned weights to capture structural information of the face. Our region layer serves as an alternative design between locally connected layers (i.e., confined kernels to individual pixels) and conventional convolution layers (i.e., shared kernels across an entire image). Unlike previous studies that solve RL and ML alternately, DRML by construction addresses both problems, allowing the two seemingly irrelevant problems to interact more directly. The complete network is end-to-end trainable, and automatically learns representations robust to variations inherent within a local region. Experiments on BP4D and DISFA benchmarks show that DRML performs the highest average F1-score and AUC within and across datasets in comparison with alternative methods.

Facial action unit (AU) detection from video has been a long-standing problem in the automated facial expression analysis. While progress has been made, accurate detection of facial AUs remains challenging due to ubiquitous sources of errors, such as inter-personal variability, pose, and low-intensity AUs. In this paper, we refer to samples causing such errors as hard samples, and the remaining as easy samples. To address learning with the hard samples, we propose the confidence preserving machine (CPM), a novel two-stage learning framework that combines multiple classifiers following an "easy-to-hard" strategy. During the training stage, CPM learns two confident classifiers. Each classifier focuses on separating easy samples of one class from all else, and thus preserves confidence on predicting each class. During the test stage, the confident classifiers provide "virtual labels" for easy test samples. Given the virtual labels, we propose a quasi-semi-supervised (QSS) learning strategy to learn a person-specific classifier. The QSS strategy employs a spatio-temporal smoothness that encourages similar predictions for samples within a spatio-temporal neighborhood. In addition, to further improve detection performance, we introduce two CPM extensions: iterative CPM that iteratively augments training samples to train the confident classifiers, and kernel CPM that kernelizes the original CPM model to promote nonlinearity. Experiments on four spontaneous data sets GFT, BP4D, DISFA, and RU-FACS illustrate the benefits of the proposed CPM models over baseline methods and the state-of-the-art semi-supervised learning and transfer learning methods.

Most action unit (AU) detection methods use one-versus-all classifiers without considering dependences between features or AUs. In this paper, we introduce a joint patch and multi-label learning (JPML) framework that models the structured joint dependence behind features, AUs, and their interplay. In particular, JPML leverages group sparsity to identify important facial patches, and learns a multi-label classifier constrained by the likelihood of co-occurring AUs. To describe such likelihood, we derive two AU relations, positive correlation and negative competition, by statistically analyzing more than 350,000 video frames annotated with multiple AUs. To the best of our knowledge, this is the first work that jointly addresses patch learning and multi-label learning for AU detection. In addition, we show that JPML can be extended to recognize holistic expressions by learning common and specific patches, which afford a more compact representation than the standard expression recognition methods. We evaluate JPML on three benchmark datasets CK+, BP4D, and GFT, using within-and cross-dataset scenarios. In four of five experiments, JPML achieved the highest averaged F1 scores in comparison with baseline and alternative methods that use either patch learning or multi-label learning alone.

We present video co-summarization, a novel perspective to video summarization that exploits visual co-occurrence across multiple videos. Motivated by the observation that important visual concepts tend to appear repeatedly across videos of the same topic, we propose to summarize a video by finding shots that co-occur most frequently across videos collected using a topic keyword. The main technical challenge is dealing with the sparsity of co-occurring patterns, out of hundreds to possibly thousands of irrelevant shots in videos being considered. To deal with this challenge, we developed a Maximal Biclique Finding (MBF) algorithm that is optimized to find sparsely co-occurring patterns, discarding less co-occurring patterns even if they are dominant in one video. Our algorithm is parallelizable with closed-form updates, thus can easily scale up to handle a large number of videos simultaneously. We demonstrate the effectiveness of our approach on motion capture and self-compiled YouTube datasets. Our results suggest that summaries generated by visual co-occurrence tend to match more closely with human generated summaries, when compared to several popular unsupervised techniques.

The face is one of the most powerful channel of nonverbal communication. The most commonly used taxonomy to describe facial behaviour is the Facial Action Coding System (FACS). FACS segments the visible effects of facial muscle activation into 30+ action units (AUs). AUs, which may occur alone and in thousands of combinations, can describe nearly all-possible facial expressions. Most existing methods for automatic AU detection treat the problem using one-vs-all classifiers and fail to exploit dependencies among AU and facial features. We introduce joint-patch and multi-label learning (JPML) to address these issues. JPML leverages group sparsity by selecting a sparse subset of facial patches while learning a multi-label classifier. In four of five comparisons on three diverse datasets, CK+, GFT, and BP4D, JPML produced the highest average F1 scores in comparison with state-of-the art.

People are inherently social. Social interaction plays an important and natural role in human behavior. Most computational methods focus on individuals alone rather than in social context. They also require labelled training data. We present an unsupervised approach to discover interpersonal synchrony, referred as to two or more persons preforming common actions in overlapping video frames or segments. For computational efficiency, we develop a branch-and-bound (B&B) approach that affords exhaustive search while guaranteeing a globally optimal solution. The proposed method is entirely general. It takes from two or more videos any multi-dimensional signal that can be represented as a histogram. We derive three novel bounding functions and provide efficient extensions, including multi-synchrony detection and accelerated search, using a warm-start strategy and parallelism. We evaluate the effectiveness of our approach in multiple databases, including human actions using the CMU Mocap dataset [1], spontaneous facial behaviors using group-formation task dataset [37] and parent-infant interaction dataset [28].

Varied sources of error contribute to the challenge of facial action unit detection. Previous approaches address specific and known sources. However, many sources are unknown. To address the ubiquity of error, we propose a Confident Preserving Machine (CPM) that follows an easy-to-hard classification strategy. During training, CPM learns two confident classifiers. A confident positive classifier separates easily identified positive samples from all else, a confident negative classifier does same for negative samples. During testing, CPM then learns a person-specific classifier using "virtual labels" provided by confident classifiers. This step is achieved using a quasi-semi-supervised (QSS) approach. Hard samples are typically close to the decision boundary, and the QSS approach disambiguates them using spatio-temporal constraints. To evaluate CPM, we compared it with a baseline single-margin classifier and state-of-the-art semi-supervised learning, transfer learning, and boosting methods in three datasets of spontaneous facial behavior. With few exceptions, CPM outperformed baseline and state-of-the art methods.

Automatic facial action unit (AFA) detection from video is a long-standing problem in facial expression analysis. Most approaches emphasize choices of features and classifiers. They neglect individual differences in target persons. People vary markedly in facial morphology (e.g., heavy versus delicate brows, smooth versus deeply etched wrinkles) and behavior. Individual differences can dramatically influence how well generic classifiers generalize to previously unseen persons. While a possible solution would be to train person-specific classifiers, that often is neither feasible nor theoretically compelling. The alternative that we propose is to personalize a generic classifier in an unsupervised manner (no additional labels for the test subjects are required). We introduce a transductive learning method, which we refer to Selective Transfer Machine (STM), to personalize a generic classifier by attenuating person-specific biases. STM achieves this effect by simultaneously learning a classifier and re-weighting the training samples that are most relevant to the test subject. To evaluate the effectiveness of STM, we compared STM to generic classifiers and to cross-domain learning methods in three major databases: CK+ [20], GEMEP-FERA [32] and RU-FACS [2]. STM outperformed generic classifiers in all.

Automatic facial Action Unit (AU) detection from video is a long-standing problem in facial expression analysis. AU detection is typically posed as a classification problem between frames or segments of positive examples and negative ones, where existing work emphasizes the use of different features or classifiers. In this paper, we propose a method called Cascade of Tasks (CoT) that combines the use of different tasks (i.e., frame, segment and transition) for AU event detection. We train CoT in a sequential manner embracing diversity, which ensures robustness and generalization to unseen data. In addition to conventional frame-based metrics that evaluate frames independently, we propose a new event-based metric to evaluate detection performance at event-level. We show how the CoT method consistently outperforms state-of-the-art approaches in both frame-based and event-based metrics, across three public datasets that differ in complexity: CK+, FERA and RU-FACS.

Accurate and efficient detection of circular objects in images is a challenging computer vision problem. Existing circular object detection methods can be broadly classified into two categories: voting based and maximum likelihood estimation (MLE) based. The former is robust to noise, however its computational complexity and memory requirement are high. On the other hand, MLE based methods (e.g., robust least squares fitting) are more computationally efficient but sensitive to noise, and can not detect multiple circles. This letter proposes Probabilistic Pairwise Voting (PPV), a fast and robust algorithm for circular object detection based on an extension of Hough Transform. The main contributions are threefold. 1) We formulate the problem of circular object detection as finding the intersection of lines in the three dimensional parameter space (i.e., center and radius of the circle). 2) We propose a probabilistic pairwise voting scheme to robustly discover circular objects under occlusion, image noise and moderate shape deformations. 3) We use a mode-finding algorithm to efficiently find multiple circular objects. We demonstrate the benefits of our approach on two real-world problems: 1) detecting circular objects in natural images, and 2) localizing iris in face images.

Rough face alignments lead to suboptimal performance of face identification systems. In this study, we present a novel approach for identifying genders from facial images without proper face alignments. Instead of using only one input for test, we generate an image set by randomly cropping out a set of image patches from a neighborhood of the face detection region. Each image set is represented as a subspace and compared with other image sets by measuring the canonical correlation between two associated subspaces. By finding an optimal discriminative transformation for all training subspaces, the proposed approach with unaligned facial images is shown to outperform the state-of-the-art methods with face alignment.