Recently, breakthroughs in video modeling have allowed for controllable camera trajectories in generated videos. However, these methods cannot be directly applied to user-provided videos that are not generated by a video model. In this paper, we present ReCapture, a method for generating new videos with novel camera trajectories from a single user-provided video. Our method allows us to re-generate the reference video, with all its existing scene motion, from vastly different angles and with cinematic camera motion. Notably, using our method we can also plausibly hallucinate parts of the scene that were not observable in the reference video. Our method works by (1) generating a noisy anchor video with a new camera trajectory using multiview diffusion models or depth-based point cloud rendering and then (2) regenerating the anchor video into a clean and temporally consistent reangled video using our proposed masked video fine-tuning technique.

We present Magic Insert, a method for dragging-and-dropping subjects from a user-provided image into a target image of a different style in a physically plausible manner while matching the style of the target image. This work formalizes the problem of style-aware drag-and-drop and presents a method for tackling it by addressing two sub-problems: style-aware personalization and realistic object insertion in stylized images. For style-aware personalization, our method first fine-tunes a pretrained text-to-image diffusion model using LoRA and learned text tokens on the subject image, and then infuses it with a CLIP representation of the target style. For object insertion, we use Bootstrapped Domain Adaption to adapt a domain-specific photorealistic object insertion model to the domain of diverse artistic styles. Overall, the method significantly outperforms traditional approaches such as inpainting. Finally, we present a dataset, SubjectPlop, to facilitate evaluation and future progress in this area. Project page: https://magicinsert.github.io/

This paper introduces a tuning-free method for both object insertion and subject-driven generation. The task involves composing an object, given multiple views, into a scene specified by either an image or text. Existing methods struggle to fully meet the task's challenging objectives: (i) seamlessly composing the object into the scene with photorealistic pose and lighting, and (ii) preserving the object's identity. We hypothesize that achieving these goals requires large scale supervision, but manually collecting sufficient data is simply too expensive. The key observation in this paper is that many mass-produced objects recur across multiple images of large unlabeled datasets, in different scenes, poses, and lighting conditions. We use this observation to create massive supervision by retrieving sets of diverse views of the same object. This powerful paired dataset enables us to train a straightforward text-to-image diffusion architecture to map the object and scene descriptions to the composited image. We compare our method, ObjectMate, with state-of-the-art methods for object insertion and subject-driven generation, using a single or multiple references. Empirically, ObjectMate achieves superior identity preservation and more photorealistic composition. Differently from many other multi-reference methods, ObjectMate does not require slow test-time tuning.

We introduce the concept of a generative infinite game, a video game that transcends the traditional boundaries of finite, hard-coded systems by using generative models. Inspired by James P. Carse's distinction between finite and infinite games, we leverage recent advances in generative AI to create Unbounded: a game of character life simulation that is fully encapsulated in generative models. Specifically, Unbounded draws inspiration from sandbox life simulations and allows you to interact with your autonomous virtual character in a virtual world by feeding, playing with and guiding it - with open-ended mechanics generated by an LLM, some of which can be emergent. In order to develop Unbounded, we propose technical innovations in both the LLM and visual generation domains. Specifically, we present: (1) a specialized, distilled large language model (LLM) that dynamically generates game mechanics, narratives, and character interactions in real-time, and (2) a new dynamic regional image prompt Adapter (IP-Adapter) for vision models that ensures consistent yet flexible visual generation of a character across multiple environments. We evaluate our system through both qualitative and quantitative analysis, showing significant improvements in character life simulation, user instruction following, narrative coherence, and visual consistency for both characters and the environments compared to traditional related approaches.

Personalization has emerged as a prominent aspect within the field of generative AI, enabling the synthesis of individuals in diverse contexts and styles, while retaining high-fidelity to their identities. However, the process of personalization presents inherent challenges in terms of time and memory requirements. Fine-tuning each personalized model needs considerable GPU time investment, and storing a personalized model per subject can be demanding in terms of storage capacity. To overcome these challenges, we propose HyperDreamBooth—a hypernetwork capable of efficiently generating a small set of personalized weights from a single image of a person. By composing these weights into the diffusion model, coupled with fast finetuning, HyperDreamBooth can generate a person's face in various contexts and styles, with high subject details while also preserving the model's crucial knowledge of diverse styles and semantic modifications. Our method achieves personalization on faces in roughly 20 seconds, 25x faster than DreamBooth and 125x faster than Textual Inversion, using as few as one reference image, with the same quality and style diversity as DreamBooth. Also our method yields a model that is 10,000x smaller than a normal DreamBooth model.

Recent advances in generative imagery have brought forth outpainting and inpainting models that can produce high-quality, plausible image content in unknown regions. However, the content these models hallucinate is necessarily inauthentic, since they are unaware of the true scene. In this work, we propose RealFill, a novel generative approach for image completion that fills in missing regions of an image with the content that should have been there. RealFill is a generative inpainting model that is personalized using only a few reference images of a scene. These reference images do not have to be aligned with the target image, and can be taken with drastically varying viewpoints, lighting conditions, camera apertures, or image styles. Once personalized, RealFill is able to complete a target image with visually compelling contents that are faithful to the original scene. We evaluate RealFill on a new image completion benchmark that covers a set of diverse and challenging scenarios, and find that it outperforms existing approaches by a large margin. Project page: https://realfill.github.io.

Recent large-scale text-guided diffusion models provide powerful image generation capabilities. Currently, a massive effort is given to enable the modification of these images using text only as means to offer intuitive and versatile editing tools. To edit a real image using these state-of-the-art tools, one must first invert the image with a meaningful text prompt into the pretrained model's domain. In this paper, we introduce an accurate inversion technique and thus facilitate an intuitive text-based modification of the image. Our proposed inversion consists of two key novel components: (i) Pivotal inversion for diffusion models. While current methods aim at mapping random noise samples to a single input image, we use a single pivotal noise vector for each timestamp and optimize around it. We demonstrate that a direct DDIM inversion is inadequate on its own, but does provide a rather good anchor for our optimization. (ii) Null-text optimization, where we only modify the unconditional textual embedding that is used for classifier-free guidance, rather than the input text embedding. This allows for keeping both the model weights and the conditional embedding intact and hence enables applying prompt-based editing while avoiding the cumbersome tuning of the model's weights. Our null-text inversion, based on the publicly available Stable Diffusion model, is extensively evaluated on a variety of images and various prompt editing, showing high-fidelity editing of real images.

Large text-to-image models achieved a remarkable leap in the evolution of AI, enabling high-quality and diverse synthesis of images from a given text prompt. However, these models lack the ability to mimic the appearance of subjects in a given reference set and synthesize novel renditions of them in different contexts. In this work, we present a new approach for “personalization” of text-to-image diffusion models. Given as input just a few images of a subject, we fine-tune a pretrained text-to-image model such that it learns to bind a unique identifier with that specific subject. Once the subject is embedded in the output domain of the model, the unique identifier can be used to synthesize novel photorealistic images of the subject contextualized in different scenes. By leveraging the semantic prior embedded in the model with a new autogenous class-specific prior preservation loss, our technique enables synthesizing the subject in diverse scenes, poses, views and lighting conditions that do not appear in the reference images. We apply our technique to several previously-unassailable tasks, including subject recontextualization, text-guided view synthesis, and artistic rendering, all while preserving the subject's key features. We also provide a new dataset and evaluation protocol for this new task of subject-driven generation. Project page: https://dreambooth.github.io/

Long exposure photography produces stunning imagery, representing moving elements in a scene with motion-blur. It is generally employed in two modalities, producing either a foreground or a background blur effect. Foreground blur images are traditionally captured on a tripod-mounted camera and portray blurred moving foreground elements, such as silky water or light trails, over a perfectly sharp background landscape. Background blur images, also called panning photography, are captured while the camera is tracking a moving subject, to produce an image of a sharp subject over a background blurred by relative motion. Both techniques are notoriously challenging and require additional equipment and advanced skills. In this paper, we describe a computational burst photography system that operates in a hand-held smartphone camera app, and achieves these effects fully automatically, at the tap of the shutter button. Our approach first detects and segments the salient subject. We track the scene motion over multiple frames and align the images in order to preserve desired sharpness and to produce aesthetically pleasing motion streaks. We capture an under-exposed burst and select the subset of input frames that will produce blur trails of controlled length, regardless of scene or camera motion velocity. We predict inter-frame motion and synthesize motion-blur to fill the temporal gaps between the input frames. Finally, we composite the blurred image with the sharp regular exposure to protect the sharpness of faces or areas of the scene that are barely moving, and produce a final high resolution and high dynamic range (HDR) photograph. Our system democratizes a capability previously reserved to professionals, and makes this creative style accessible to most casual photographers.

Using only a model that was trained to predict where people look at images, and no additional training data, we can produce a range of powerful editing effects for reducing distraction in images. Given an image and a mask specifying the region to edit, we backpropagate through a state-of-the-art saliency model to parameterize a differentiable editing operator, such that the saliency within the masked region is reduced. We demonstrate several operators, including: a recoloring operator, which learns to apply a color transform that camouflages and blends distractors into their surroundings; a warping operator, which warps less salient image regions to cover distractors, gradually collapsing objects into themselves and effectively removing them (an effect akin to inpainting); a GAN operator, which uses a semantic prior to fully replace image regions with plausible, less salient alternatives. The resulting effects are consistent with cognitive research on the human visual system (e.g., since color mismatch is salient, the recoloring operator learns to harmonize objects' colors with their surrounding to reduce their saliency). And importantly, all effects are achieved under a zero-shot learning scenario, solely through the guidance of the pretrained saliency model, with no supervised data of the effects. We present results on a variety of natural images and conduct a perceptual study to evaluate and validate the changes in viewers' eye-gaze between the original images and our edited results. Project Webpage: https://deep-saliency-prior.github.io/

We introduce MyStyle, a personalized deep generative prior trained with a few shots of an individual. MyStyle allows to reconstruct, enhance and edit images of a specific person, such that the output is faithful to the person's key facial characteristics. Given a small reference set of portrait images of a person (~ 100), we tune the weights of a pretrained StyleGAN face generator to form a local, low-dimensional, personalized manifold in the latent space. We show that this manifold constitutes a personalized region that spans latent codes associated with diverse portrait images of the individual. Moreover, we demonstrate that we obtain a personalized generative prior, and propose a unified approach to apply it to various ill-posed image enhancement problems, such as inpainting and super-resolution, as well as semantic editing. Using the personalized generative prior we obtain outputs that exhibit high-fidelity to the input images and are also faithful to the key facial characteristics of the individual in the reference set. We demonstrate our method with fair-use images of numerous widely recognizable individuals for whom we have the prior knowledge for a qualitative evaluation of the expected outcome. We evaluate our approach against few-shots baselines and show that our personalized prior, quantitatively and qualitatively, outperforms state-of-the-art alternatives.

Taking photographs in low light using a mobile phone is challenging and rarely produces pleasing results. Aside from the physical limits imposed by read noise and photon shot noise, these cameras are typically handheld, have small apertures and sensors, use mass-produced analog electronics that cannot easily be cooled, and are commonly used to photograph subjects that move, like children and pets. In this paper we describe a system for capturing clean, sharp, colorful photographs in light as low as 0.3 lux, where human vision becomes monochromatic and indistinct. To permit handheld photography without flash illumination, we capture, align, and combine multiple frames. Our system employs "motion metering", which uses an estimate of motion magnitudes (whether due to handshake or moving objects) to identify the number of frames and the per-frame exposure times that together minimize both noise and motion blur in a captured burst. We combine these frames using robust alignment and merging techniques that are specialized for high-noise imagery. To ensure accurate colors in such low light, we employ a learning-based auto white balancing algorithm. To prevent the photographs from looking like they were shot in daylight, we use tone mapping techniques inspired by illusionistic painting: increasing contrast, crushing shadows to black, and surrounding the scene with darkness. All of these processes are performed using the limited computational resources of a mobile device. Our system can be used by novice photographers to produce shareable pictures in a few seconds based on a single shutter press, even in environments so dim that humans cannot see clearly.

Shallow depth-of-field is commonly used by photographers to isolate a subject from a distracting background. However, standard cell phone cameras cannot produce such images optically, as their short focal lengths and small apertures capture nearly all-in-focus images. We present a system to computationally synthesize shallow depth-of-field images with a single mobile camera and a single button press. If the image is of a person, we use a person segmentation network to separate the person and their accessories from the background. If available, we also use dense dual-pixel auto-focus hardware, effectively a 2-sample light field with an approximately 1 millimeter baseline, to compute a dense depth map. These two signals are combined and used to render a defocused image. Our system can process a 5.4 megapixel image in 4 seconds on a mobile phone, is fully automatic, and is robust enough to be used by non-experts. The modular nature of our system allows it to degrade naturally in the absence of a dual-pixel sensor or a human subject.

We address the problem of stereoscopic content generation from light fields using multi-perspective imaging. Our proposed method takes as input a light field and a target disparity map, and synthesizes a stereoscopic image pair by selecting light rays that fulfill the given target disparity constraints. We formulate this as a variational convex optimization problem. Compared to previous work, our method makes use of multi-view input to composite the new view with occlusions and disocclusions properly handled, does not require any correspondence information such as scene depth, is free from undesirable artifacts such as grid bias or image distortion, and is more efficiently solvable. In particular, our method is about ten times more memory efficient than the previous art, and is capable of processing higher resolution input. This is essential to make the proposed method practically applicable to realistic scenarios where HD content is standard. We demonstrate the effectiveness of our method experimentally.

We present a solution for generating high-quality stereo panoramas at mega pixel resolutions. While previous approaches introduced the basic principles, we show that those techniques do not generalise well to today's high image resolutions and lead to disturbing visual artefacts. As our first contribution, we describe the necessary correction steps and a compact representation for the input images in order to achieve a highly accurate approximation to the required ray space. Our second contribution is a flow-based up sampling of the available input rays which effectively resolves known aliasing issues like stitching artefacts. The required rays are generated on the fly to perfectly match the desired output resolution, even for small numbers of input images. In addition, the up sampling is real-time and enables direct interactive control over the desired stereoscopic depth effect. In combination, our contributions allow the generation of stereoscopic panoramas at high output resolutions that are virtually free of artefacts such as seams, stereo discontinuities, vertical parallax and other mono-/stereoscopic shape distortions. Our process is robust, and other types of multiperspective panoramas, such as linear panoramas, can also benefit from our contributions. We show various comparisons and high-resolution results.

In this paper we propose a novel method to compress video content based on image retargeting. First, a saliency map is extracted from the video frames either automatically or according to user input. Next, nonlinear image scaling is performed which assigns a higher pixel count to salient image regions and fewer pixels to non-salient regions. The non-linearly downscaled images can then be compressed using existing compression techniques and decoded and upscaled at the receiver. To this end we introduce a non-uniform antialiasing technique that significantly improves the image resampling quality. The overall process is complementary to existing compression methods and can be seamlessly incorporated into existing pipelines. We compare our method to JPEG 2000 and H.264/AVC-10 and show that, at the cost of visual quality in non-salient image regions, our method achieves a significant improvement of the visual quality of salient image regions in terms of Structural Similarity (SSIM) and Peak Signal-to-Noise-Ratio (PSNR) quality measures, in particular for scenarios with high compression ratios.

This paper describes a method for scene reconstruction of complex, detailed environments from 3D light fields. Densely sampled light fields in the order of 10 9 light rays allow us to capture the real world in unparalleled detail, but efficiently processing this amount of data to generate an equally detailed reconstruction represents a significant challenge to existing algorithms. We propose an algorithm that leverages coherence in massive light fields by breaking with a number of established practices in image-based reconstruction. Our algorithm first computes reliable depth estimates specifically around object boundaries instead of interior regions, by operating on individual light rays instead of image patches. More homogeneous interior regions are then processed in a fine-to-coarse procedure rather than the standard coarse-to-fine approaches. At no point in our method is any form of global optimization performed. This allows our algorithm to retain precise object contours while still ensuring smooth reconstructions in less detailed areas. While the core reconstruction method handles general unstructured input, we also introduce a sparse representation and a propagation scheme for reliable depth estimates which make our algorithm particularly effective for 3D input, enabling fast and memory efficient processing of "Gigaray light fields" on a standard GPU. We show dense 3D reconstructions of highly detailed scenes, enabling applications such as automatic segmentation and image-based rendering, and provide an extensive evaluation and comparison to existing image-based reconstruction techniques.

Saliency estimation has become a valuable tool in image processing. Yet, existing approaches exhibit considerable variation in methodology, and it is often difficult to attribute improvements in result quality to specific algorithm properties. In this paper we reconsider some of the design choices of previous methods and propose a conceptually clear and intuitive algorithm for contrast-based saliency estimation. Our algorithm consists of four basic steps. First, our method decomposes a given image into compact, perceptually homogeneous elements that abstract unnecessary detail. Based on this abstraction we compute two measures of contrast that rate the uniqueness and the spatial distribution of these elements. From the element contrast we then derive a saliency measure that produces a pixel-accurate saliency map which uniformly covers the objects of interest and consistently separates fore- and background. We show that the complete contrast and saliency estimation can be formulated in a unified way using high-dimensional Gaussian filters. This contributes to the conceptual simplicity of our method and lends itself to a highly efficient implementation with linear complexity. In a detailed experimental evaluation we analyze the contribution of each individual feature and show that our method outperforms all state-of-the-art approaches.

Millions of surveillance cameras record video around the clock, producing huge video archives. Even when a video archive is known to include critical activities, finding them is like finding a needle in a haystack, making the archive almost worthless. Two main approaches were proposed to address this problem: action recognition and video summarization. Methods for automatic detection of activities still face problems in many scenarios. The video synopsis approach to video summarization is very effective, but may produce confusing summaries by the simultaneous display of multiple activities.A new methodology for the generation of short and coherent video summaries is presented, based on clustering of similar activities. Objects with similar activities are easy to watch simultaneously, and outliers can be spotted instantly. Clustered synopsis is also suitable for efficient creation of ground truth data.

Geometric rearrangement of images includes operations such as image retargeting, inpainting, or object rearrangement. Each such operation can be characterized by a shiftmap: the relative shift of every pixel in the output image from its source in an input image. We describe a new representation of these operations as an optimal graph labeling, where the shift-map represents the selected label for each output pixel. Two terms are used in computing the optimal shift-map: (i) A data term which indicates constraints such as the change in image size, object rearrangement, a possible saliency map, etc. (ii) A smoothness term, minimizing the new discontinuities in the output image caused by discontinuities in the shift-map. This graph labeling problem can be solved using graph cuts. Since the optimization is global and discrete, it outperforms state of the art methods in most cases. Efficient hierarchical solutions for graph-cuts are presented, and operations on 1M images can take only a few seconds.

The amount of captured video is growing with the increased numbers of video cameras, especially the increase of millions of surveillance cameras that operate 24 hours a day. Since video browsing and retrieval is time consuming, most captured video is never watched or examined. Video synopsis is an effective tool for browsing and indexing of such a video. It provides a short video representation, while preserving the essential activities of the original video. The activity in the video is condensed into a shorter period by simultaneously showing multiple activities, even when they originally occurred at different times. The synopsis video is also an index into the original video by pointing to the original time of each activity. Video Synopsis can be applied to create a synopsis of an endless video streams, as generated by webcams and by surveillance cameras. It can address queries like "Show in one minute the synopsis of this camera broadcast during the past day''. This process includes two major phases: (i) An online conversion of the endless video stream into a database of objects and activities (rather than frames). (ii) A response phase, generating the video synopsis as a response to the user's query.

The world is covered with millions of Webcams, many transmit everything in their field of view over the Internet 24 hours a day. A Web search finds public webcams in airports, intersections, classrooms, parks, shops, ski resorts, and more. Even more private surveillance cameras cover many private and public facilities. Webcams are an endless resource, but most of the video broadcast will be of little interest due to lack of activity. We propose to generate a short video that will be a synopsis of an endless video streams, generated by webcams or surveillance cameras. We would like to address queries like "I would like to watch in one minute the highlights of this camera broadcast during the past day". The process includes two major phases: (i) An online conversion of the video stream into a database of objects and activities (rather than frames), (ii) A response phase, generating the video synopsis as a response to the user's query. To include maximum information in a short synopsis we simultaneously show activities that may have happened at different times. The synopsis video can also be used as an index into the original video stream.

This paper explores the manipulation of time in video editing, enabling to control the chronological time of events. These time manipulations include slowing down (or postponing) some dynamic events while speeding up (or advancing) others. When a video camera scans a scene, aligning all the events to a single time interval will result in a panoramic movie. Time manipulations are obtained by first constructing an aligned space-time volume from the input video, and then sweeping a continuous 2D slice (time front) through that volume, generating a new sequence of images. For dynamic scenes, aligning the input video frames poses an important challenge. We propose to align dynamic scenes using a new notion of "dynamics constancy", which is more appropriate for this task than the traditional assumption of "brightness constancy". Another challenge is to avoid visual seams inside moving objects and other visual artifacts resulting from sweeping the space-time volumes with time fronts of arbitrary geometry. To avoid such artifacts, we formulate the problem of finding optimal time front geometry as one of finding a minimal cut in a 4D graph, and solve it using max-flow methods.

The power of video over still images is the ability to represent dynamic activities. But video browsing and retrieval are inconvenient due to inherent spatio-temporal redundancies, where some time intervals may have no activity, or have activities that occur in a small image region. Video synopsis aims to provide a compact video representation, while preserving the essential activities of the original video. We present dynamic video synopsis, where most of the activity in the video is condensed by simultaneously showing several actions, even when they originally occurred at different times. For example, we can create a "stroboscopic movie", where multiple dynamic instances of a moving object are played simultaneously. This is an extension of the still stroboscopic picture. Previous approaches for video abstraction addressed mostly the temporal redundancy by selecting representative key-frames or time intervals. In dynamic video synopsis the activity is shifted into a significantly shorter period, in which the activity is much denser. Video examples can be found online in http://www.vision.huji.ac.il/synopsis

With the limited field of view of human vision, our perception of most scenes is built over time while our eyes are scanning the scene. In the case of static scenes, this process can be modeled by panoramic mosaicing: stitching together images into a panoramic view. Can a dynamic scene, scanned by a video camera, be represented with a dynamic panoramic video even though different regions were visible at different times? In this paper, we explore time flow manipulation in video, such as the creation of new videos in which events that occurred at different times are displayed simultaneously. More general changes in the time flow are also possible, which enable re-scheduling the order of dynamic events in the video, for example. We generate dynamic mosaics by sweeping the aligned space-time volume of the input video by a time front surface and generating a sequence of time slices in the process. Various sweeping strategies and different time front evolutions manipulate the time flow in the video, enabling many unexplored and powerful effects, such as panoramic movies.

An omnistereo panorama consists of a pair of panoramic images, where one panorama is for the left eye and another panorama is for the right eye. The panoramic stereo pair provides a stereo sensation up to a full 360 degrees. Omnistereo panoramas can be constructed by mosaicing images from a single rotating camera. This approach also enables the control of stereo disparity, giving larger baselines for faraway scenes, and a smaller baseline for closer scenes. Capturing panoramic omnistereo images with a rotating camera makes it impossible to capture dynamic scenes at video rates and limits omnistereo imaging to stationary scenes. We present two possibilities for capturing omnistereo panoramas using optics without any moving parts. A special mirror is introduced such that viewing the scene through this mirror creates the same rays as those used with the rotating cameras. The lens used for omnistereo panorama is also introduced, together with the design of the mirror. Omnistereo panoramas can also be rendered by computer graphics methods to represent virtual environments.

A panorama for visual stereo consists of a pair of panoramic images, where one panorama is for the left eye, and another panorama is for the right eye. A panoramic stereo pair provides a stereo sensation lip to a full 360 degrees. A stereo panorama cannot be photographed by two omnidirectional cameras from two viewpoints. It is normally constructed by mosaicing together images from a rotating stereo pair, or from a single moving camera. Capturing stereo panoramic images by a rotating camera makes it impossible to capture dynamic scenes at video rates, and limits stereo panoramic imaging to stationary scenes. This paper presents two possibilities for capturing stereo panoramic images using optics, without any moving parts. A special mirror is introduced such that viewing the scene through this mirror creates the same rays as those used with the rotating cameras. Such a mirror enables the capture of stereo panoramic movies with a regular video camera. A lens for stereo panorama is also introduced. The designs of the mirror and of the lens are based on curves whose caustic is a circle.