With models getting stronger, evaluations have grown more complex, testing multiple skills in one benchmark and even in the same instance at once. However, skill-wise performance is obscured when inspecting aggregate accuracy, under-utilizing the rich signal modern benchmarks contain. We propose an automatic approach to recover the underlying skills relevant for any evaluation instance, by way of inspecting model-generated {\em rationales}. After validating the relevance of rationale-parsed skills and inferring skills for $46$k instances over $12$ benchmarks, we observe many skills to be common across benchmarks, resulting in the curation of hundreds of \emph{skill-slices} (i.e. sets of instances testing a common skill). Inspecting accuracy over these slices yields novel insights on model trade-offs: e.g., compared to GPT-4o and Claude 3.5 Sonnet, on average, Gemini 1.5 Pro is $18\%$ more accurate in \emph{computing molar mass}, but $19\\%$ less accurate in \emph{applying constitutional law}, despite the overall accuracies of the three models differing by a mere $0.4\\%$. Furthermore, we demonstrate the practical utility of our approach by showing that insights derived from skill slice analysis can generalize to held-out instances: when routing each instance to the model strongest on the relevant skills, we see a $3\\%$ accuracy improvement over our $12$ dataset corpus. Our skill-slices and framework open a new avenue in model evaluation, leveraging skill-specific analyses to unlock a more granular and actionable understanding of model capabilities.

Large language models (LLMs) and vision-language models (VLMs) have demonstrated remarkable performance across a wide range of tasks and domains. Despite this promise, spatial understanding and reasoning—a fundamental component of human cognition—remains under-explored. We propose SpatialEval, a novel benchmark that covers diverse aspects of spatial reasoning such as relationship understanding, navigation, and counting. We conduct a comprehensive evaluation of competitive language and vision-language models. Our findings reveal several counter-intuitive insights that have been overlooked in the literature: (1) Spatial reasoning poses significant challenges where competitive models can fall behind random guessing; (2) Despite additional visual input, VLMs often under-perform compared to their LLM counterparts; (3) When both textual and visual information is available, multi-modal language models become less reliant on visual information if sufficient textual clues are provided. Additionally, we demonstrate that leveraging redundancy between vision and text can significantly enhance model performance. We hope our study will inform the development of multimodal models to improve spatial intelligence and further close the gap with human intelligence. Our code is available at https://github.com/jiayuww/SpatialEval.

It is very time consuming to create datasets for training computer vision models. An emerging alternative is to use synthetic data, but if the synthetic data is not similar enough to the real data, the performance is typically below that of training with real data. Thus using synthetic data still requires a large amount of time, money, and skill as one needs to author the data carefully. In this paper, we seek to understand which aspects of this authoring process are most critical. We present an analysis of which factors of variation between simulated and real data are most important. We capture images of YCB objects to create a novel YCB-Real dataset. We then create a novel synthetic "digital twin" dataset, YCB-Synthetic, which matches the YCB-Real dataset and includes variety of artifacts added to the synthetic data. We study the affects of these artifacts on our dataset and two existing published datasets on two different computer vision tasks: object detection and instance segmentation. We provide an analysis of the cost-benefit trade-offs between artist time for fixing artifacts and trained model accuracy. We plan to release this dataset (images and 3D assets) so they can be further used by the community. Link to dataset1

Building an interactive AI assistant that can perceive, reason, and collaborate with humans in the real world has been a long-standing pursuit in the AI community. This work is part of a broader research effort to develop intelligent agents that can interactively guide humans through performing tasks in the physical world. As a first step in this direction, we introduce HoloAssist, a large-scale egocentric human interaction dataset, where two people collaboratively complete physical manipulation tasks. The task performer executes the task while wearing a mixed-reality headset that captures seven synchronized data streams. The task instructor watches the performer’s egocentric video in real time and guides them verbally. By augmenting the data with action and conversational annotations and observing the rich behaviors of various participants, we present key insights into how human assistants correct mistakes, intervene in the task completion procedure, and ground their instructions to the environment. HoloAssist spans 166 hours of data captured by 350 unique instructor-performer pairs. Furthermore, we construct and present benchmarks on mistake detection, intervention type prediction, and hand forecasting, along with detailed analysis. We expect HoloAssist will provide an important resource for building AI assistants that can fluidly collaborate with humans in the real world. Data can be downloaded at https://holoassist.github.io/.

A critical object detection task is finetuning an existing model to detect novel objects, but the standard workflow requires bounding box annotations which are time-consuming and expensive to collect. Weakly supervised object detection (WSOD) offers an appealing alternative, where object detectors can be trained using image-level labels. However, the practical application of current WSOD models is limited, as they only operate at small data scales and require multiple rounds of training and refinement. To address this, we propose the Weakly Supervised Detection Transformer, which enables efficient knowledge transfer from a large-scale pretraining dataset to WSOD finetuning on hundreds of novel objects. Additionally, we leverage pretrained knowledge to improve the multiple instance learning (MIL) framework often used in WSOD methods. Our experiments show that our approach outperforms previous state-of-the-art models on large-scale novel object detection datasets, and our scaling study reveals that class quantity is more important than image quantity for WSOD pretraining.

Contrastive learning relies on an assumption that positive pairs contain related views that share certain underlying information about an instance, e.g., patches of an image or co-occurring multimodal signals of a video. What if this assumption is violated? The literature suggests that contrastive learning produces suboptimal representations in the presence of noisy views, e.g., false positive pairs with no apparent shared information. In this work, we pro-pose a new contrastive loss function that is robust against noisy views. We provide rigorous theoretical justifications by showing connections to robust symmetric losses for noisy binary classification and by establishing a new contrastive bound for mutual information maximization based on the Wasserstein distance measure. The proposed loss is completely modality-agnostic and a simple drop-in replacement for the InfoNCE loss, which makes it easy to apply to ex-isting contrastive frameworks. We show that our approach provides consistent improvements over the state-of-the-art on image, video, and graph contrastive learning bench-marks that exhibit a variety of real-world noise patterns.

Visual attention helps achieve robust perception under noise, corruption, and distribution shifts in human vision, which are areas where modern neural networks still fall short. We present VARS, Visual Attention from Recurrent Sparse reconstruction, a new attention formulation built on two prominent features of the human visual attention mechanism: recurrency and sparsity. Related features are grouped together via recurrent connections between neurons, with salient objects emerging via sparse regularization. VARS adopts an attractor network with recurrent connections that converges toward a stable pattern over time. Network layers are represented as ordinary differential equations (ODEs), formulating attention as a recurrent attractor network that equivalently optimizes the sparse reconstruction of input using a dictionary of “templates” encoding underlying patterns of data. We show that self-attention is a special case of VARS with a single-step optimization and no sparsity constraint. VARS can be readily used as a replacement for self-attention in popular vision transformers, consistently improving their robustness across various benchmarks.

Recent work has shown that the structure of convolutional neural networks (CNNs) induces a strong prior that favors natural images. This prior, known as a deep image prior (DIP), is an effective regularizer in inverse problems such as image denoising and inpainting. We extend the concept of the DIP to depth images. Given color images and noisy and incomplete target depth maps, we optimize a randomly-initialized CNN model to reconstruct a depth map restored by virtue of using the CNN network structure as a prior combined with a view-constrained photo-consistency loss. This loss is computed using images from a geometrically calibrated camera from nearby viewpoints. We apply this deep depth prior for inpainting and refining incomplete and noisy depth maps within both binocular and multi-view stereo pipelines. Our quantitative and qualitative evaluation shows that our refined depth maps are more accurate and complete, and after fusion, produces dense 3D models of higher quality.

The ability to detect and classify rare occurrences in images has important applications - for example, counting rare and endangered species when studying biodiversity, or detecting infrequent traffic scenarios that pose a danger to self-driving cars. Few-shot learning is an open problem: current computer vision systems struggle to categorize objects they have seen only rarely during training, and collecting a sufficient number of training examples of rare events is often challenging and expensive, and sometimes outright impossible. We explore in depth an approach to this problem: complementing the few available training images with ad-hoc simulated data.Our testbed is animal species classification, which has a real-world long-tailed distribution. We present two natural world simulators, and analyze the effect of different axes of variation in simulation, such as pose, lighting, model, and simulation method, and we prescribe best practices for efficiently incorporating simulated data for real-world performance gain. Our experiments reveal that synthetic data can considerably reduce error rates for classes that are rare, that as the amount of simulated data is increased, accuracy on the target class improves, and that high variation of simulated data provides maximum performance gain.

We present a system for converting a fully panoramic (360 degree) video into a normal field-of-view (NFOV) hyperlapse for an optimal viewing experience. Our system exploits visual saliency and semantics to non-uniformly sample in space and time for generating hyperlapses. In addition, users can optionally choose objects of interest for customizing the hyperlapses. We first stabilize an input 360 degree video by smoothing the rotation between adjacent frames and then compute regions of interest and saliency scores. An initial hyperlapse is generated by optimizing the saliency and motion smoothness followed by the saliency-aware frame selection. We further smooth the result using an efficient 2D video stabilization approach that adaptively selects the motion model to generate the final hyperlapse. We validate the design of our system by showing results for a variety of scenes and comparing against the state-of-the-art method through a large-scale user study.

Cinemagraphs are a compelling way to convey dynamic aspects of a scene. In these media, dynamic and still elements are juxtaposed to create an artistic and narrative experience. Creating a high-quality, aesthetically pleasing cinemagraph requires isolating objects in a semantically meaningful way and then selecting good start times and looping periods for those objects to minimize visual artifacts (such a tearing). To achieve this, we present a new technique that uses object recognition and semantic segmentation as part of an optimization method to automatically create cinemagraphs from videos that are both visually appealing and semantically meaningful. Given a scene with multiple objects, there are many cinemagraphs one could create. Our method evaluates these multiple candidates and presents the best one, as determined by a model trained to predict human preferences in a collaborative way. We demonstrate the effectiveness of our approach with multiple results and a user study.

Drones equipped with cameras are emerging as a powerful tool for large-scale aerial 3D scanning, but existing automatic flight planners do not exploit all available information about the scene, and can therefore produce inaccurate and incomplete 3D models. We present an automatic method to generate drone trajectories, such that the imagery acquired during the flight will later produce a high-fidelity 3D model. Our method uses a coarse estimate of the scene geometry to plan camera trajectories that: (1) cover the scene as thoroughly as possible; (2) encourage observations of scene geometry from a diverse set of viewing angles; (3) avoid obstacles; and (4) respect a user-specified flight time budget. Our method relies on a mathematical model of scene coverage that exhibits an intuitive diminishing returns property known as submodularity. We leverage this property extensively to design a trajectory planning algorithm that reasons globally about the non-additive coverage reward obtained across a trajectory, jointly with the cost of traveling between views. We evaluate our method by using it to scan three large outdoor scenes, and we perform a quantitative evaluation using a photorealistic video game simulator.

Emerging uses of imaging technology for consumers cover a wide range of application areas from health to interaction techniques; however, typical cameras primarily transduce light from the visible spectrum into only three overlapping components of the spectrum: red, blue, and green. In contrast, hyperspectral imaging breaks down the electromagnetic spectrum into more narrow components and expands coverage beyond the visible spectrum. While hyperspectral imaging has proven useful as an industrial technology, its use as a sensing approach has been fragmented and largely neglected by the UbiComp community. We explore an approach to make hyperspectral imaging easier and bring it closer to the end-users. HyperCam provides a low-cost implementation of a multispectral camera and a software approach that automatically analyzes the scene and provides a user with an optimal set of images that try to capture the salient information of the scene. We present a number of use-cases that demonstrate HyperCam's usefulness and effectiveness.

Logging food and calorie intake has been shown to facilitate weight management. Unfortunately, current food logging methods are time-consuming and cumbersome, which limits their effectiveness. To address this limitation, we present an automated computer vision system for logging food and calorie intake using images. We focus on the "restaurant" scenario, which is often a challenging aspect of diet management. We introduce a key insight that addresses this problem specifically: restaurant plates are often both nutritionally and visually consistent across many servings. This insight provides a path to robust calorie estimation from a single RGB photograph: using a database of known food items together with restaurant-specific classifiers, calorie estimation can be achieved through identification followed by calorie lookup. As demonstrated on a challenging Menu-Match dataset and an existing third party dataset, our approach outperforms previous computer vision methods and a commercial calorie estimation app. Our Menu-Match dataset of realistic restaurant meals is made publicly available.

Long videos can be played much faster than real-time by recording only one frame per second or by dropping all but one frame each second, i.e., by creating a timelapse. Unstable hand-held moving videos can be stabilized with a number of recently described methods. Unfortunately, creating a stabilized timelapse, or hyperlapse , cannot be achieved through a simple combination of these two methods. Two hyperlapse methods have been previously demonstrated: one with high computational complexity and one requiring special sensors. We present an algorithm for creating hyperlapse videos that can handle significant high-frequency camera motion and runs in real-time on HD video. Our approach does not require sensor data, thus can be run on videos captured on any camera. We optimally select frames from the input video that best match a desired target speed-up while also resulting in the smoothest possible camera motion. We evaluate our approach using several input videos from a range of cameras and compare these results to existing methods.

Movie directors and cinematographers impart style onto video using techniques that are learned through years of experience: camera movement, framing, color, lighting, etc. Without this experience and expensive equipment, it is very difficult to control stylistic aspects of a video. We introduce a novel approach for post-hoc editing of one specific aspect of cinematography -- camera motion style -- via an equalizer-like set of controls that manipulates the power spectra of a video's apparent motion path. We explore free manipulation of apparent camera motion as well as the transfer of motion styles from an example video to a new video to create a wide range of stylistic variations. We report on a user study confirming the ability of non-expert users to create motion styles.

It is often desirable to evaluate images quality with a perceptually relevant measure that does not require a reference image. Recent approaches to this problem use human provided quality scores with machine learning to learn a measure. The biggest hurdles to these efforts are: 1) the difficulty of generalizing across diverse types of distortions and 2) collecting the enormity of human scored training data that is needed to learn the measure. We present a new blind image quality measure that addresses these difficulties by learning a robust, nonlinear kernel regression function using a rectifier neural network. The method is pre-trained with unlabeled data and fine-tuned with labeled data. It generalizes across a large set of images and distortion types without the need for a large amount of labeled data. We evaluate our approach on two benchmark datasets and show that it not only outperforms the current state of the art in blind image quality estimation, but also outperforms the state of the art in non-blind measures. Furthermore, we show that our semi-supervised approach is robust to using varying amounts of labeled data.

Given a short video we create a representation that captures a spectrum of looping videos with varying levels of dynamism, ranging from a static image to a highly animated loop. In such a progressively dynamic video, scene liveliness can be adjusted interactively using a slider control. Applications include background images and slideshows, where the desired level of activity may depend on personal taste or mood. The representation also provides a segmentation of the scene into independently looping regions, enabling interactive local adjustment over dynamism. For a landscape scene, this control might correspond to selective animation and deanimation of grass motion, water ripples, and swaying trees. Converting arbitrary video to looping content is a challenging research problem. Unlike prior work, we explore an optimization in which each pixel automatically determines its own looping period. The resulting nested segmentation of static and dynamic scene regions forms an extremely compact representation.

We present a touch-free interface for viewing large imagery on mobile devices. In particular, we focus on viewing paradigms for 360 degree panoramas, parallax image sequences, and long multi-perspective panoramas. We describe a sensor fusion methodology that combines face tracking using a front-facing camera with gyroscope data to produce a robust signal that defines the viewer's 3D position relative to the display. The gyroscopic data provides both low-latency feedback and allows extrapolation of the face position beyond the the field-of-view of the front-facing camera. We also demonstrate a hybrid position and rate control that uses the viewer's 3D position to drive exploration of very large image spaces. We report on the efficacy of the hybrid control vs. position only control through a user study.

We explore creating ""cliplets"", a form of visual media that juxtaposes still image and video segments, both spatially and temporally, to expressively abstract a moment. Much as in ""cinemagraphs"", the tension between static and dynamic elements in a cliplet reinforces both aspects, strongly focusing the viewer's attention. Creating this type of imagery is challenging without professional tools and training. We develop a set of idioms, essentially spatiotemporal mappings, that characterize cliplet elements, and use these idioms in an interactive system to quickly compose a cliplet from ordinary handheld video. One difficulty is to avoid artifacts in the cliplet composition without resorting to extensive manual input. We address this with automatic alignment, looping optimization and feathering, simultaneous matting and compositing, and Laplacian blending. A key user-interface challenge is to provide affordances to define the parameters of the mappings from input time to output time while maintaining a focus on the cliplet being created. We demonstrate the creation of a variety of cliplet types. We also report on informal feedback as well as a more structured survey of users.

The restoration of a blurry or noisy image is commonly performed with a MAP estimator, which maximizes a posterior probability to reconstruct a clean image from a degraded image. A MAP estimator, when used with a sparse gradient image prior, reconstructs piecewise smooth images and typically removes textures that are important for visual realism. We present an alternative deconvolution method called iterative distribution reweighting (IDR) which imposes a global constraint on gradients so that a reconstructed image should have a gradient distribution similar to a reference distribution. In natural images, a reference distribution not only varies from one image to another, but also within an image depending on texture. We estimate a reference distribution directly from an input image for each texture segment. Our algorithm is able to restore rich mid-frequency textures. A large-scale user study supports the conclusion that our algorithm improves the visual realism of reconstructed images compared to those of MAP estimators.

We describe a unified framework for generating a single high-quality still image ("snapshot") from a short video clip. Our system allows the user to specify the desired operations for creating the output image, such as super resolution, noise and blur reduction, and selection of best focus. It also provides a visual summary of activity in the video by incorporating saliency-based objectives in the snapshot formation process. We show examples on a number of different video clips to illustrate the utility and flexibility of our system.

We address the effects of noise in low-light images in this paper. Color images are usually captured by a sensor with a color filter array (CFA). This requires a demosaicing process to generate a full color image. The captured images typically have low signal-to-noise ratio, and the demosaicing step further corrupts the image, which we show to be the leading cause of visually objectionable random noise patterns (splotches). To avoid this problem, we propose a combined framework of denoising and demosaicing, where we use information about the image inferred in the denoising step to perform demosaicing. Our experiments show that such a framework results in sharper low-light images that are devoid of splotches and other noise artifacts.

It is often desirable to evaluate an image based on its quality. For many computer vision applications, a perceptually meaningful measure is the most relevant for evaluation; however, most commonly used measure do not map well to human judgements of image quality. A further complication of many existing image measure is that they require a reference image, which is often not available in practice. In this paper, we present a “blind” image quality measure, where potentially neither the groundtruth image nor the degradation process are known. Our method uses a set of novel low-level image features in a machine learning framework to learn a mapping from these features to subjective image quality scores. The image quality features stem from natural image measure and texture statistics. Experiments on a standard image quality benchmark dataset shows that our method outperforms the current state of art.

In image restoration tasks, a heavy-tailed gradient distribution of natural images has been extensively exploited as an image prior. Most image restoration algorithms impose a sparse gradient prior on the whole image, reconstructing an image with piecewise smooth characteristics. While the sparse gradient prior removes ringing and noise artifacts, it also tends to remove mid-frequency textures, degrading the visual quality. We can attribute such degradations to imposing an incorrect image prior. The gradient profile in fractal-like textures, such as trees, is close to a Gaussian distribution, and small gradients from such regions are severely penalized by the sparse gradient prior. To address this issue, we introduce an image restoration algorithm that adapts the image prior to the underlying texture. We adapt the prior to both low-level local structures as well as mid-level textural characteristics. Improvements in visual quality is demonstrated on deconvolution and denoising tasks.

In this paper, we show how to generate a sharp panorama from a set of motion-blurred video frames. Our technique is based on joint global motion estimation and multi-frame deblurring. It also automatically computes the duty cycle of the video, namely the percentage of time between frames that is actually exposure time. The duty cycle is necessary for allowing the blur kernels to be accurately extracted and then removed. We demonstrate our technique on a number of videos.

We present a deblurring algorithm that uses a hardware attachment coupled with a natural image prior to deblur images from consumer cameras. Our approach uses a combination of inexpensive gyroscopes and accelerometers in an energy optimization framework to estimate a blur function from the camera's acceleration and angular velocity during an exposure. We solve for the camera motion at a high sampling rate during an exposure and infer the latent image using a joint optimization. Our method is completely automatic, handles per-pixel, spatially-varying blur, and out-performs the current leading image-based methods. Our experiments show that it handles large kernels -- up to at least 100 pixels, with a typical size of 30 pixels. We also present a method to perform "ground-truth" measurements of camera motion blur. We use this method to validate our hardware and deconvolution approach. To the best of our knowledge, this is the first work that uses 6 DOF inertial sensors for dense, per-pixel spatially-varying image deblurring and the first work to gather dense ground-truth measurements for camera-shake blur.

We describe a framework for improving the quality of personal photos by using a person's favorite photographs as examples. We observe that the majority of a person's photographs include the faces of a photographer's family and friends and often the errors in these photographs are the most disconcerting. We focus on correcting these types of images and use common faces across images to automatically perform both global and face-specific corrections. Our system achieves this by using face detection to align faces between “good” and “bad” photos such that properties of the good examples can be used to correct a bad photo. These “personal” photos provide strong guidance for a number of operations and, as a result, enable a number of high-quality image processing operations. We illustrate the power and generality of our approach by presenting a novel deblurring algorithm, and we show corrections that perform sharpening, superresolution, in-painting of over- and underexposured regions, and white-balancing.

Image blur and noise are difficult to avoid in many situations and can often ruin a photograph. We present a novel image deconvolution algorithm that deblurs and denoises an image given a known shift-invariant blur kernel. Our algorithm uses local color statistics derived from the image as a constraint in a unified framework that can be used for deblurring, denoising, and upsampling. A pixel's color is required to be a linear combination of the two most prevalent colors within a neighborhood of the pixel. This two-color prior has two major benefits: it is tuned to the content of the particular image and it serves to decouple edge sharpness from edge strength. Our unified algorithm for deblurring and denoising out-performs previous methods that are specialized for these individual applications. We demonstrate this with both qualitative results and extensive quantitative comparisons that show that we can out-perform previous methods by approximately 1 to 3 DB.

This paper presents a simple yet practical 3-D modeling method for recovering surface shape and reflectance from a set of images. We attach a point light source to a hand-held camera to add a photometric constraint to the multi-view stereo problem. Using the photometric constraint, we simultaneously solve for shape, surface normal, and reflectance. Unlike prior approaches, we formulate the problem using realistic assumptions of a near light source, non-Lambertian surfaces, perspective camera model, and the presence of ambient lighting. The effectiveness of the proposed method is verified using simulated and real-world scenes.

Image blur is caused by a number of factors such as motion, defocus, capturing light over the non-zero area of the aperture and pixel, the presence of anti-aliasing filters on a camera sensor, and limited sensor resolution. We present an algorithm that estimates non-parametric, spatially-varying blur functions (i.e., point-spread functions or PSFs) at subpixel resolution from a single image. Our method handles blur due to defocus, slight camera motion, and inherent aspects of the imaging system. Our algorithm can be used to measure blur due to limited sensor resolution by estimating a sub-pixel, super-resolved PSF even for in-focus images. It operates by predicting a ldquosharprdquo version of a blurry input image and uses the two images to solve for a PSF. We handle the cases where the scene content is unknown and also where a known printed calibration target is placed in the scene. Our method is completely automatic, fast, and produces accurate results.

Occlusion is a significant challenge for many tracking algorithms. Most current methods can track through transient occlusion, but cannot handle significant extended occlusion when the object's trajectory may change significantly. We present a method to track a 3D object through significant occlusion using multiple nearby cameras (e.g., a camera array). When an occluder and object are at different depths, different parts of the object are visible or occluded in each view due to parallax. By aggregating across these views, the method can track even when any individual camera observes very little of the target object. Implementation- wise, the methods are straightforward and build upon established single-camera algorithms. They do not require explicit modeling or reconstruction of the scene and enable tracking in complex, dynamic scenes with moving cameras. Analysis of accuracy and robustness shows that these methods are successful when upwards of '70% of the object is occluded in every camera view. To the best of our knowledge, this system is the first capable of tracking in the presence of such significant occlusion.

We address the problem of reconstructing the 3-D shape of a Lambertian surface from multiple images acquired as an object rotates under distant and possibly varying illumination. Using camera projection matrices estimated from point correspondences across views, the algorithm computes a dense correspondence map by minimizing a multi-ocular photometric constraint. Once correspondence across views is established, photometric stereo is applied to estimate a surface normal field and 3-D surface. Conceptually, the algorithm merges multi-view stereo and photometric stereo and uses aspects of both methods to recover shape. The method is straightforward to implement and relies on established principles from the two stereo methods. We empirically validate the method on images of a number of objects and show that it outperforms previous methods.

We present an algorithm and a system for high-quality natural video matting using a camera array. The system uses high frequencies present in natural scenes to compute mattes by creating a synthetic aperture image that is focused on the foreground object, which reduces the variance of pixels reprojected from the foreground while increasing the variance of pixels reprojected from the background. We modify the standard matting equation to work directly with variance measurements and show how these statistics can be used to construct a trimap that is later upgraded to an alpha matte. The entire process is completely automatic, including an automatic method for focusing the synthetic aperture image on the foreground object and an automatic method to compute the trimap and the alpha matte. The proposed algorithm is very efficient and has a per-pixel running time that is linear in the number of cameras. Our current system runs at several frames per second, and we believe that it is the first system capable of computing high-quality alpha mattes at near real-time rates without the use of active illumination or special backgrounds.

The advent of inexpensive digital image sensors and the ability to create photographs that combine information from a number of sensed images are changing the way we think about photography. In this paper, we describe a unique array of 100 custom video cameras that we have built, and we summarize our experiences using this array in a range of imaging applications. Our goal was to explore the capabilities of a system that would be inexpensive to produce in the future. With this in mind, we used simple cameras, lenses, and mountings, and we assumed that processing large numbers of images would eventually be easy and cheap. The applications we have explored include approximating a conventional single center of projection video camera with high performance along one or more axes, such as resolution, dynamic range, frame rate, and/or large aperture, and using multiple cameras to approximate a video camera with a large synthetic aperture. This permits us to capture a video light field, to which we can apply spatiotemporal view interpolation algorithms in order to digitally simulate time dilation and camera motion. It also permits us to create video sequences using custom non-uniform synthetic apertures.