Scene coordinate regression (SCR) has established itself as a promising learning-based approach to visual relocalization. After mere minutes of scene-specific training, SCR models estimate camera poses of query images with high accuracy. Still, SCR methods fall short of the generalization capabilities of more classical feature-matching approaches. When imaging conditions of query images, such as lighting or viewpoint, are too different from the training views, SCR models fail. Failing to generalize is an inherent limitation of previous SCR frameworks, since their training objective is to encode the training views in the weights of the coordinate regressor itself. The regressor essentially overfits to the training views, by design. We propose to separate the coordinate regressor and the map representation into a generic transformer and a scene-specific map code. This separation allows us to pre-train the transformer on tens of thousands of scenes. More importantly, it allows us to train the transformer to generalize from mapping images to unseen query images during pre-training. We demonstrate on multiple challenging relocalization datasets that our method, ACE-G, leads to significantly increased robustness while keeping the computational footprint attractive

Neural Radiance Fields (NeRFs) are emerging as a ubiquitous scene representation that allows for novel view synthesis. Increasingly, NeRFs will be shareable with other people. Before sharing a NeRF, though, it might be desirable to remove personal information or unsightly objects. Such removal is not easily achieved with the current NeRF editing frameworks. We propose a framework to remove objects from a NeRF representation created from an RGBD sequence. Our NeRF inpainting method leverages recent work in 2D image inpainting and is guided by a userprovided mask. Our algorithm is underpinned by a confidence based view selection procedure. It chooses which of the individual 2D inpainted images to use in the creation of the NeRF, so that the resulting inpainted NeRF is 3D consistent. We show that our method for NeRF editing is effective for synthesizing plausible inpaintings in a multi-view coherent manner, outperforming competing methods. We validate our approach by proposing a new and still-challenging dataset for the task of NeRF inpainting.

Visual re-localization means using a single image as input to estimate the camera’s location and orientation relative to a pre-recorded environment. The highest-scoring methods are “structure-based,” and need the query camera’s intrinsics as an input to the model, with careful geometric optimization. When intrinsics are absent, methods vie for accuracy by making various other assumptions. This yields fairly good localization scores, but the models are “narrow” in some way, e.g., requiring costly test-time computations, or depth sensors, or multiple query frames. In contrast, our proposed method makes few special assumptions, and is fairly lightweight in training and testing.Our pose regression network learns from only relative poses of training scenes. For inference, it builds a graph connecting the query image to training counterparts and uses a graph neural network (GNN) with image representations on nodes and image-pair representations on edges. By efficiently passing messages between them, both representation types are refined to produce a consistent camera pose estimate. We validate the effectiveness of our approach on both standard indoor (7-Scenes) and outdoor (Cambridge Landmarks) camera re-localization benchmarks. Our relative pose regression method matches the accuracy of absolute pose regression networks, while retaining the relative-pose models’ test-time speed and ability to generalize to non-training scenes.

Understanding the shape of a scene from a single color image is a formidable computer vision task. However, most methods aim to predict the geometry of surfaces that are visible to the camera, which is of limited use when planning paths for robots or augmented reality agents. Such agents can only move when grounded on a traversable surface, which we define as the set of classes which humans can also walk over, such as grass, footpaths and pavement. Models which predict beyond the line of sight often parameterize the scene with voxels or meshes, which can be expensive to use in machine learning frameworks. We introduce a model to predict the geometry of both visible and occluded traversable surfaces, given a single RGB image as input. We learn from stereo video sequences, using camera poses, per-frame depth and semantic segmentation to form training data, which is used to supervise an image-to-image network. We train models from the KITTI driving dataset, the indoor Matterport dataset, and from our own casually captured stereo footage. We find that a surprisingly low bar for spatial coverage of training scenes is required. We validate our algorithm against a range of strong baselines, and include an assessment of our predictions for a path-planning task.

We present RELATE, a model that learns to generate physically plausible scenes and videos of multiple interacting objects. Similar to other generative approaches, RELATE is trained end-to-end on raw, unlabeled data. RELATE combines an object-centric GAN formulation with a model that explicitly accounts for correlations between individual objects. This allows the model to generate realistic scenes and videos from a physically-interpretable parameterization. Furthermore, we show that modeling the object correlation is necessary to learn to disentangle object positions and identity. We find that RELATE is also amenable to physically realistic scene editing and that it significantly outperforms prior art in object-centric scene generation in both synthetic (CLEVR, ShapeStacks) and real-world data (cars). In addition, in contrast to state-of-the-art methods in object-centric generative modeling, RELATE also extends naturally to dynamic scenes and generates videos of high visual fidelity. Source code, datasets and more results are available at http://geometry.cs.ucl.ac.uk/projects/2020/relate/.

While the basic laws of Newtonian mechanics are well understood, explaining a physical scenario still requires manually modeling the problem with suitable equations and estimating the associated parameters. In order to be able to leverage the approximation capabilities of artificial intelligence techniques in such physics related contexts, researchers have handcrafted relevant states, and then used neural networks to learn the state transitions using simulation runs as training data. Unfortunately, such approaches are unsuited for modeling complex real-world scenarios, where manually authoring relevant state spaces tend to be tedious and challenging. In this work, we investigate if neural networks can implicitly learn physical states of real-world mechanical processes only based on visual data while internally modeling non-homogeneous environment and in the process enable long-term physical extrapolation. We develop a recurrent neural network architecture for this task and also characterize resultant uncertainties in the form of evolving variance estimates. We evaluate our setup, to extrapolate motion of rolling ball(s) on bowls of varying shape and orientation, and on arbitrary heightfields using only images as input. We report significant improvements over existing image-based methods both in terms of accuracy of predictions and complexity of scenarios; and report competitive performance with approaches that, unlike us, assume access to internal physical states.

Next generation smart and augmented reality systems demand a computational understanding of monocular footage that captures humans in physical spaces to reveal plausible object arrangements and human-object interactions. Despite recent advances, both in scene layout and human motion analysis, the above setting remains challenging to analyze due to regular occlusions that occur between objects and human motions. We observe that the interaction between object arrangements and human actions is often strongly correlated, and hence can be used to help recover from these occlusions. We present iMapper, a data-driven method to identify such human-object interactions and utilize them to infer layouts of occluded objects. Starting from a monocular video with detected 2D human joint positions that are potentially noisy and occluded, we first introduce the notion of interaction-saliency as space-time snapshots where informative human-object interactions happen. Then, we propose a global optimization to retrieve and fit interactions from a database to the detected salient interactions in order to best explain the input video. We extensively evaluate the approach, both quantitatively against manually annotated ground truth and through a user study, and demonstrate that iMapper produces plausible scene layouts for scenes with medium to heavy occlusion. Code and data are available on the project page.

Curved folded surfaces, given their ability to produce elegant freeform shapes by folding flat sheets etched with curved creases, hold a special place in computational Origami. Artists and designers have proposed a wide variety of different fold patterns to create a range of interesting surfaces. The creative process, design, as well as fabrication is usually only concerned with the static surface that emerges once folding has completed. Folding such patterns, however, is difficult as multiple creases have to be folded simultaneously to obtain a properly folded target shape. We introduce string actuated curved folded surfaces that can be shaped by pulling a network of strings, thus, vastly simplifying the process of creating such surfaces and making the folding motion an integral part of the design. Technically, we solve the problem of which surface points to string together and how to actuate them by locally expressing a desired folding path in the space of isometric shape deformations in terms of novel string actuation modes. We demonstrate the validity of our approach by computing string actuation networks for a range of well-known crease patterns and testing their effectiveness on physical prototypes. All the examples in this article can be downloaded for personal use from http://geometry.cs.ucl.ac.uk/projects/2017/string-actuated/.

Collision sequences are commonly used in games and entertainment to add drama and excitement. Authoring even two body collisions in the real world can be difficult, as one has to get timing and the object trajectories to be correctly synchronized. After tedious trial-and-error iterations, when objects can actually be made to collide, then they are difficult to capture in 3D. In contrast, synthetically generating plausible collisions is difficult as it requires adjusting different collision parameters (e.g.,object mass ratio, coefficient of restitution,etc.) and appropriate initial parameters. We present SMASH to directlyread offappropriate collision parameters directly from raw input video recordings. Technically we enable this by utilizing laws of rigid body collision to regularize the problem of lifting 2D trajectories to a physically valid 3D reconstruction of the collision. The reconstructed sequences can then be modified and combined to easily author novel and plausible collisions. We evaluate our system on a range of synthetic scenes and demonstrate the effectiveness of our method by accurately reconstructing several complex real world collision events.

With the proliferation of acquisition devices, gathering massive volumes of 3D data is now easy. Processing such large masses of pointclouds, however, remains a challenge. This is particularly a problem for raw scans with missing data, noise, and varying sampling density. In this work, we present a simple, scalable, yet powerful data reconstruction algorithm. We focus on reconstruction of man-made scenes as regular arrangements of planes (RAP), thereby selecting both local plane-based approximations along with their global inter-plane relations. We propose a novel selection formulation to directly balance between data fitting and the simplicity of the resulting arrangement of extracted planes. The main technical contribution is a formulation that allows less-dominant orientations to still retain their internal regularity, and not become overwhelmed and regularized by the dominant scene orientations. We evaluate our approach on a variety of complex 2D and 3D pointclouds, and demonstrate the advantages over existing alternative methods.

Missing data due to occlusion is a key challenge in 3D acquisition, particularly in cluttered man-made scenes. Such partial information about the scenes limits our ability to analyze and understand them. In this work we abstract such environments as collections of cuboids and hallucinate geometry in the occluded regions by globally analyzing the physical stability of the resultant arrangements of the cuboids. Our algorithm extrapolates the cuboids into the un-seen regions to infer both their corresponding geometric attributes (e.g., size, orientation) and how the cuboids topologically interact with each other (e.g., touch or fixed). The resultant arrangement provides an abstraction for the underlying structure of the scene that can then be used for a range of common geometry processing tasks. We evaluate our algorithm on a large number of test scenes with varying complexity, validate the results on existing benchmark datasets, and demonstrate the use of the recovered cuboid-based structures towards object retrieval, scene completion, etc.