Reconstructing precise camera poses and floor plan layouts from wide-baseline RGB panoramas is a difficult and unsolved problem. We introduce BADGR, a novel diffusion model that jointly performs reconstruction and bundle adjustment (BA) to refine poses and layouts from a coarse state, using 1D floor boundary predictions from dozens of sparsely captured images. Unlike guided diffusion models, BADGR is conditioned on dense per-column outputs from a single-step Levenberg Marquardt (LM) optimizer and is trained to predict camera and wall positions, while minimizing reprojection errors for view consistency. The objective of layout generation from denoising diffusion process complements BA optimization by providing additional learned layout-structural constraints on top of the co-visible features across images. These constraints help BADGR make plausible guesses about spatial relationships, which constrain the pose graph, such as wall adjacency and collinearity, while also learning to mitigate errors from dense boundary observations using global context. BADGR trains exclusively on 2D floor plans, simplifying data acquisition, enabling robust augmentation, and supporting a variety of input densities. Our experiments validate our method, which significantly outperforms the state-of-the-art pose and floor plan layout reconstruction with different input densities. Visit project website at: https://badgr-diffusion.github.io.

Rendering scenes observed in a monocular video from novel viewpoints is a challenging problem. For static scenes the community has studied both scene-specific optimization techniques, which optimize on every test scene, and generalized techniques, which only run a deep net forward pass on a test scene. In contrast, for dynamic scenes, scene-specific optimization techniques exist, but, to our best knowledge, there is currently no generalized method for dynamic novel view synthesis from a given monocular video. To explore whether generalized dynamic novel view synthesis from monocular videos is possible today, we establish an analysis framework based on existing techniques and work toward the generalized approach. We find a pseudo-generalized process without scene-specific \emph{appearance} optimization is possible, but geometrically and temporally consistent depth estimates are needed. Despite no scene-specific appearance optimization, the pseudo-generalized approach improves upon some scene-specific methods.For more information see project page at https://xiaoming-zhao.github.io/projects/pgdvs.

Real world images often have highly imbalanced content density. Some areas are very uniform, e.g., large patches of blue sky, while other areas are scattered with many small objects. Yet, the commonly used successive grid downsampling strategy in convolutional deep networks treats all areas equally. Hence, small objects are represented in very few spatial locations, leading to worse results in tasks such as segmentation. Intuitively, retaining more pixels representing small objects during downsampling helps to preserve important information. To achieve this, we propose AutoFocusFormer (AFF), a local-attention transformer image recognition backbone, which performs adaptive down-sampling by learning to retain the most important pixels for the task. Since adaptive downsampling generates a set of pixels irregularly distributed on the image plane, we abandon the classic grid structure. Instead, we develop a novel point-based local attention block, facilitated by a balanced clustering module and a learnable neighborhood merging module, which yields representations for our point-based versions of state-of-the-art segmentation heads. Experiments show that our AutoFocusFormer (AFF) improves significantly over baseline models of similar sizes.

Dense 3D reconstruction from RGB images traditionally assumes static camera pose estimates. This assumption has endured, even as recent works have increasingly focused on real-time methods for mobile devices. However, the assumption of a fixed pose for each image does not hold for online execution: poses from real-time SLAM are dynamic and may be updated following events such as bundle adjustment and loop closure. This has been addressed in the RGB-D setting, by de-integrating past views and re-integrating them with updated poses, but it remains largely untreated in the RGB-only setting. We formalize this problem to define the new task of dense online reconstruction from dynamically-posed images. To support further research, we introduce a dataset called LivePose1containing the dynamic poses from a SLAM system running on Scan-Net [6]. We select three recent reconstruction systems and apply a framework based on de-integration to adapt each one to the dynamic-pose setting. In addition, we propose a novel, non-linear de-integration module that learns to remove stale scene content. We show that responding to pose updates is critical for high-quality reconstruction, and that our de-integration framework is an effective solution.

Recent works on 3D reconstruction from posed images [17], [23], [24] have demonstrated that direct inference of scene-level 3D geometry without test-time optimization is feasible using deep neural networks, showing remarkable promise and high efficiency. However, the reconstructed geometry, typically represented as a 3D truncated signed distance function (TSDF), is often coarse without fine geometric details. To address this problem, we propose three effective solutions for improving the fidelity of inference-based 3D reconstructions. We first present a resolution-agnostic TSDF supervision strategy to provide the network with a more accurate learning signal during training, avoiding the pitfalls of TSDF interpolation seen in previous work. We then introduce a depth guidance strategy using multi-view depth estimates to enhance the scene representation and recover more accurate surfaces. Finally, we develop a novel architecture for the final layers of the network, conditioning the output TSDF prediction on high-resolution image features in addition to coarse voxel features, enabling sharper reconstruction of fine details. Our method, FineRecon1, produces smooth and highly accurate reconstructions, showing significant improvements across multiple depth and 3D reconstruction metrics.

We study the problem of novel view synthesis from sparse source observations of a scene comprised of 3D objects. We propose a simple yet effective approach that is neither continuous nor implicit, challenging recent trends on view synthesis. Our approach explicitly encodes observations into a volumetric representation that enables amortized rendering. We demonstrate that although continuous radiance field representations have gained a lot of attention due to their expressive power, our simple approach obtains comparable or even better novel view reconstruction quality comparing with state-of-the-art baselines [49] while increasing rendering speed by over 400x. Our model is trained in a category-agnostic manner and does not require scene-specific optimization. Therefore, it is able to generalize novel view synthesis to object categories not seen during training. In addition, we show that with our simple formulation, we can use view synthesis as a self-supervision signal for efficient learning of 3D geometry without explicit 3D supervision.

We propose a framework for learning neural scene representations directly from images, without 3D supervision. Our key insight is that 3D structure can be imposed by ensuring that the learned representation transforms like a real 3D scene. Specifically, we introduce a loss which enforces equivariance of the scene representation with respect to 3D transformations. Our formulation allows us to infer and render scenes in real time while achieving comparable results to models requiring minutes for inference. In addition, we introduce two challenging new datasets for scene representation and neural rendering, including scenes with complex lighting and backgrounds. Through experiments, we show that our model achieves compelling results on these datasets as well as on standard ShapeNet benchmarks.

We propose an algorithm to predict room layout from a single image that generalizes across panoramas and perspective images, cuboid layouts and more general layouts (e.g. "L"-shape room). Our method operates directly on the panoramic image, rather than decomposing into perspective images as do recent works. Our network architecture is similar to that of RoomNet [15], but we show improvements due to aligning the image based on vanishing points, predicting multiple layout elements (corners, boundaries, size and translation), and fitting a constrained Manhattan layout to the resulting predictions. Our method compares well in speed and accuracy to other existing work on panoramas, achieves among the best accuracy for perspective images, and can handle both cuboid-shaped and more general Manhattan layouts.

We present ExprGen, a system to automatically generate 3D stylized character expressions from humans in a perceptually valid and geometrically consistent manner. Our multi-stage deep learning system utilizes the latent variables of human and character expression recognition convolutional neural networks to control a 3D animated character rig. This end-to-end system takes images of human faces and generates the character rig parameters that best match the human's facial expression. ExprGen generalizes to multiple characters, and allows expression transfer between characters in a semi-supervised manner. Qualitative and quantitative evaluation of our method based on Mechanical Turk tests show the high perceptual accuracy of our expression transfer results.

Imagining what a proposed home remodel might look like without actually performing it is challenging. We present an image-based remodeling methodology that allows real-time photorealistic visualization during both the modeling and remodeling process of a home interior. Large-scale edits, like removing a wall or enlarging a window, are performed easily and in real time, with realistic results. Our interface supports the creation of concise, parameterized, and constrained geometry, as well as remodeling directly from within the photographs. Real-time texturing of modified geometry is made possible by precomputing view-dependent textures for all faces that are potentially visible to each original camera viewpoint, blending multiple viewpoints and hole-filling when necessary. The resulting textures are stored and accessed efficiently enabling intuitive real-time realistic visualization, modeling, and editing of the building interior.

We present an interactive system for efficiently extracting foreground objects from a video. We extend previous min-cut based image segmentation techniques to the domain of video with four new contributions. We provide a novel painting-based user interface that allows users to easily indicate the foreground object across space and time. We introduce a hierarchical mean-shift preprocess in order to minimize the number of nodes that min-cut must operate on. Within the min-cut we also define new local cost functions to augment the global costs defined in earlier work. Finally, we extend 2D alpha matting methods designed for images to work with 3D video volumes. We demonstrate that our matting approach preserves smoothness across both space and time. Our interactive video cutout system allows users to quickly extract foreground objects from video sequences for use in a variety of applications including compositing onto new backgrounds and NPR cartoon style rendering.

We describe an interactive, computer-assisted framework for combining parts of a set of photographs into a single composite picture, a process we call "digital photomontage." Our framework makes use of two techniques primarily: graph-cut optimization, to choose good seams within the constituent images so that they can be combined as seamlessly as possible; and gradient-domain fusion, a process based on Poisson equations, to further reduce any remaining visible artifacts in the composite. Also central to the framework is a suite of interactive tools that allow the user to specify a variety of high-level image objectives, either globally across the image, or locally through a painting-style interface. Image objectives are applied independently at each pixel location and generally involve a function of the pixel values (such as "maximum contrast") drawn from that same location in the set of source images. Typically, a user applies a series of image objectives iteratively in order to create a finished composite. The power of this framework lies in its generality; we show how it can be used for a wide variety of applications, including "selective composites" (for instance, group photos in which everyone looks their best), relighting, extended depth of field, panoramic stitching, clean-plate production, stroboscopic visualization of movement, and time-lapse mosaics.

The common meeting is an integral part of everyday life for most workgroups. However, due to travel, time, or other constraints, people are often not able to attend all the meetings they need to. Teleconferencing and recording of meetings can address this problem. In this paper we describe a system that provides these features, as well as a user study evaluation of the system. The system uses a variety of capture devices (a novel 360° camera, a whiteboard camera, an overview camera, and a microphone array) to provide a rich experience for people who want to participate in a meeting from a distance. The system is also combined with speaker clustering, spatial indexing, and time compression to provide a rich experience for people who miss a meeting and want to watch it afterward.