Matching local geometric features on real-world depth images is a challenging task due to the noisy, low-resolution, and incomplete nature of 3D scan data. These difficulties limit the performance of current state-of-art methods, which are typically based on histograms over geometric properties. In this paper, we present 3DMatch, a data-driven model that learns a local volumetric patch descriptor for establishing correspondences between partial 3D data. To amass training data for our model, we propose a self-supervised feature learning method that leverages the millions of correspondence labels found in existing RGB-D reconstructions. Experiments show that our descriptor is not only able to match local geometry in new scenes for reconstruction, but also generalize to different tasks and spatial scales (e.g. instance-level object model alignment for the Amazon Picking Challenge, and mesh surface correspondence). Results show that 3DMatch consistently outperforms other state-of-the-art approaches by a significant margin. Code, data, benchmarks, and pre-trained models are available online at http://3dmatch.cs.princeton.edu.

3D context has been shown to be extremely important for scene understanding, yet very little research has been done on integrating context information with deep neural network architectures. This paper presents an approach to embed 3D context into the topology of a neural network trained to perform holistic scene understanding. Given a depth image depicting a 3D scene, our network aligns the observed scene with a predefined 3D scene template, and then reasons about the existence and location of each object within the scene template. In doing so, our model recognizes multiple objects in a single forward pass of a 3D convolutional neural network, capturing both global scene and local object information simultaneously. To create training data for this 3D network, we generate partially synthetic depth images which are rendered by replacing real objects with a repository of CAD models of the same object category1. Extensive experiments demonstrate the effectiveness of our algorithm compared to the state of the art.

Robot warehouse automation has attracted significant interest in recent years, perhaps most visibly in the Amazon Picking Challenge (APC) [1]. A fully autonomous warehouse pick-and-place system requires robust vision that reliably recognizes and locates objects amid cluttered environments, self-occlusions, sensor noise, and a large variety of objects. In this paper we present an approach that leverages multiview RGB-D data and self-supervised, data-driven learning to overcome those difficulties. The approach was part of the MIT-Princeton Team system that took 3rd- and 4th-place in the stowing and picking tasks, respectively at APC 2016. In the proposed approach, we segment and label multiple views of a scene with a fully convolutional neural network, and then fit pre-scanned 3D object models to the resulting segmentation to get the 6D object pose. Training a deep neural network for segmentation typically requires a large amount of training data. We propose a self-supervised method to generate a large labeled dataset without tedious manual segmentation. We demonstrate that our system can reliably estimate the 6D pose of objects under a variety of scenarios. All code, data, and benchmarks are available at http://apc.cs.princeton.edu/

We focus on the task of amodal 3D object detection in RGB-D images, which aims to produce a 3D bounding box of an object in metric form at its full extent. We introduce Deep Sliding Shapes, a 3D ConvNet formulation that takes a 3D volumetric scene from a RGB-D image as input and outputs 3D object bounding boxes. In our approach, we propose the first 3D Region Proposal Network (RPN) to learn objectness from geometric shapes and the first joint Object Recognition Network (ORN) to extract geometric features in 3D and color features in 2D. In particular, we handle objects of various sizes by training an amodal RPN at two different scales and an ORN to regress 3D bounding boxes. Experiments show that our algorithm outperforms the state-of-the-art by 13.8 in mAP and is 200× faster than the original Sliding Shapes.

Basic Linear Algebra Subprograms (BLAS) are a set of low level linear algebra kernels widely adopted by applications involved with the deep learning and scientific computing. The massive and economic computing power brought forth by the emerging GPU architectures drives interest in implementation of compute-intensive level 3 BLAS on multi-GPU systems. In this paper, we investigate existing multi-GPU level 3 BLAS and present that 1) issues, such as the improper load balancing, inefficient communication, insufficient GPU stream level concurrency and data caching, impede current implementations from fully harnessing heterogeneous computing resources; 2) and the inter-GPU Peer-to-Peer(P2P) communication remains unexplored. We then present BLASX: a highly optimized multi-GPU level-3 BLAS. We adopt the concepts of algorithms-by-tiles treating a matrix tile as the basic data unit and operations on tiles as the basic task. Tasks are guided with a dynamic asynchronous runtime, which is cache and locality aware. The communication cost under BLASX becomes trivial as it perfectly overlaps communication and computation across multiple streams during asynchronous task progression. It also takes the current tile cache scheme one step further by proposing an innovative 2-level hierarchical tile cache, taking advantage of inter-GPU P2P communication. As a result, linear speedup is observable with BLASX under multi-GPU configurations; and the extensive benchmarks demonstrate that BLASX consistently outperforms the related leading industrial and academic implementations such as cuBLAS-XT, SuperMatrix, MAGMA.

Although RGB-D sensors have enabled major break-throughs for several vision tasks, such as 3D reconstruction, we have not attained the same level of success in high-level scene understanding. Perhaps one of the main reasons is the lack of a large-scale benchmark with 3D annotations and 3D evaluation metrics. In this paper, we introduce an RGB-D benchmark suite for the goal of advancing the state-of-the-arts in all major scene understanding tasks. Our dataset is captured by four different sensors and contains 10,335 RGB-D images, at a similar scale as PASCAL VOC. The whole dataset is densely annotated and includes 146,617 2D polygons and 64,595 3D bounding boxes with accurate object orientations, as well as a 3D room layout and scene category for each image. This dataset enables us to train data-hungry algorithms for scene-understanding tasks, evaluate them using meaningful 3D metrics, avoid overfitting to a small testing set, and study cross-sensor bias.

3D shape is a crucial but heavily underutilized cue in today's computer vision systems, mostly due to the lack of a good generic shape representation. With the recent availability of inexpensive 2.5D depth sensors (e.g. Microsoft Kinect), it is becoming increasingly important to have a powerful 3D shape representation in the loop. Apart from category recognition, recovering full 3D shapes from view-based 2.5D depth maps is also a critical part of visual understanding. To this end, we propose to represent a geometric 3D shape as a probability distribution of binary variables on a 3D voxel grid, using a Convolutional Deep Belief Network. Our model, 3D ShapeNets, learns the distribution of complex 3D shapes across different object categories and arbitrary poses from raw CAD data, and discovers hierarchical compositional part representation automatically. It naturally supports joint object recognition and shape completion from 2.5D depth maps, and it enables active object recognition through view planning. To train our 3D deep learning model, we construct ModelNet - a large-scale 3D CAD model dataset. Extensive experiments show that our 3D deep representation enables significant performance improvement over the-state-of-the-arts in a variety of tasks.

This paper describes an automatic algorithm for global alignment of LiDAR data collected with Google Street View cars in urban environments. The problem is challenging because global pose estimation techniques (GPS) do not work well in city environments with tall buildings, and local tracking techniques (integration of inertial sensors, structure-from-motion, etc.) provide solutions that drift over long ranges, leading to solutions where data collected over wide ranges is warped and misaligned by many meters. Our approach to address this problem is to extract “semantic features” with object detectors (e.g., for facades, poles, cars, etc.) that can be matched robustly at different scales, and thus are selected for different iterations of an ICP algorithm. We have implemented an all-to-all, non-rigid, global alignment based on this idea that provides better results than alternatives during experiments with data from large regions of New York, San Francisco, Paris, and Rome.

Today, there are two major paradigms for vision-based autonomous driving systems: mediated perception approaches that parse an entire scene to make a driving decision, and behavior reflex approaches that directly map an input image to a driving action by a regressor. In this paper, we propose a third paradigm: a direct perception approach to estimate the affordance for driving. We propose to map an input image to a small number of key perception indicators that directly relate to the affordance of a road/traffic state for driving. Our representation provides a set of compact yet complete descriptions of the scene to enable a simple controller to drive autonomously. Falling in between the two extremes of mediated perception and behavior reflex, we argue that our direct perception representation provides the right level of abstraction. To demonstrate this, we train a deep Convolutional Neural Network using recording from 12 hours of human driving in a video game and show that our model can work well to drive a car in a very diverse set of virtual environments. We also train a model for car distance estimation on the KITTI dataset. Results show that our direct perception approach can generalize well to real driving images. Source code and data are available on our project website.

Global bundle adjustment usually converges to a non-zero residual and produces sub-optimal camera poses for local areas, which leads to loss of details for high- resolution reconstruction. Instead of trying harder to optimize everything globally, we argue that we should live with the non-zero residual and adapt the camera poses to local areas. To this end, we propose a segment-based approach to readjust the camera poses locally and improve the reconstruction for fine geometry details. The key idea is to partition the globally optimized structure from motion points into well-conditioned segments for re-optimization, reconstruct their geometry individually, and fuse everything back into a consistent global model. This significantly reduces severe propagated errors and estimation biases caused by the initial global adjustment. The results on several datasets demonstrate that this approach can significantly improve the reconstruction accuracy, while maintaining the consistency of the 3D structure between segments.

When glancing at a magazine, or browsing the Internet, we are continuously exposed to photographs. Despite this overflow of visual information, humans are extremely good at remembering thousands of pictures along with some of their visual details. But not all images are equal in memory. Some stick in our minds while others are quickly forgotten. In this paper, we focus on the problem of predicting how memorable an image will be. We show that memorability is an intrinsic and stable property of an image that is shared across different viewers, and remains stable across delays. We introduce a database for which we have measured the probability that each picture will be recognized after a single view. We analyze a collection of image features, labels, and attributes that contribute to making an image memorable, and we train a predictor based on global image descriptors. We find that predicting image memorability is a task that can be addressed with current computer vision techniques. While making memorable images is a challenging task in visualization, photography, and education, this work is a first attempt to quantify this useful property of images.

We propose a novel linear method to match cuboids in indoor scenes using RGBD images from Kinect. Beyond depth maps, these cuboids reveal important structures of a scene. Instead of directly fitting cuboids to 3D data, we first construct cuboid candidates using super pixel pairs on a RGBD image, and then we optimize the configuration of the cuboids to satisfy the global structure constraints. The optimal configuration has low local matching costs, small object intersection and occlusion, and the cuboids tend to project to a large region in the image, the number of cuboids is optimized simultaneously. We formulate the multiple cuboid matching problem as a mixed integer linear program and solve the optimization efficiently with a branch and bound method. The optimization guarantees the global optimal solution. Our experiments on the Kinect RGBD images of a variety of indoor scenes show that our proposed method is efficient, accurate and robust against object appearance variations, occlusions and strong clutter.

We significantly extrapolate the field of view of a photograph by learning from a roughly aligned, wide-angle guide image of the same scene category. Our method can extrapolate typical photos into complete panoramas. The extrapolation problem is formulated in the shift-map image synthesis framework. We analyze the self-similarity of the guide image to generate a set of allowable local transformations and apply them to the input image. Our guided shift-map method reserves to the scene layout of the guide image when extrapolating a photograph. While conventional shift-map methods only support translations, this is not expressive enough to characterize the self-similarity of complex scenes. Therefore we additionally allow image transformations of rotation, scaling and reflection. To handle this increase in complexity, we introduce a hierarchical graph optimization method to choose the optimal transformation at each output pixel. We demonstrate our approach on a variety of indoor, outdoor, natural, and man-made scenes.

In this paper, we develop a generative model to describe the layouts of outdoor scenes - the spatial configuration of regions. Specifically, the layout of an image is represented as a composite of regions, each associated with a semantic topic. At the heart of this model is a novel stochastic process called Spatial Topic Process, which generates a spatial map of topics from a set of coupled Gaussian processes, thus allowing the distributions of topics to vary continuously across the image plane. A key aspect that distinguishes this model from previous ones consists in its capability of capturing dependencies across both locations and topics while allowing substantial variations in the layouts. We demonstrate the practical utility of the proposed model by testing it on scene classification, semantic segmentation, and layout hallucination.

We present a data-driven method for building dense 3D reconstructions using a combination of recognition and multi-view cues. Our approach is based on the idea that there are image patches that are so distinctive that we can accurately estimate their latent 3D shapes solely using recognition. We call these patches shape anchors, and we use them as the basis of a multi-view reconstruction system that transfers dense, complex geometry between scenes. We "anchor" our 3D interpretation from these patches, using them to predict geometry for parts of the scene that are relatively ambiguous. The resulting algorithm produces dense reconstructions from stereo point clouds that are sparse and noisy, and we demonstrate it on a challenging dataset of real-world, indoor scenes.

Despite significant progress, tracking is still considered to be a very challenging task. Recently, the increasing popularity of depth sensors has made it possible to obtain reliable depth easily. This may be a game changer for tracking, since depth can be used to prevent model drift and handle occlusion. We also observe that current tracking algorithms are mostly evaluated on a very small number of videos collected and annotated by different groups. The lack of a reasonable size and consistently constructed benchmark has prevented a persuasive comparison among different algorithms. In this paper, we construct a unified benchmark dataset of 100 RGBD videos with high diversity, propose different kinds of RGBD tracking algorithms using 2D or 3D model, and present a quantitative comparison of various algorithms with RGB or RGBD input. We aim to lay the foundation for further research in both RGB and RGBD tracking, and our benchmark is available at http://tracking.cs.princeton.edu.

Existing scene understanding datasets contain only a limited set of views of a place, and they lack representations of complete 3D spaces. In this paper, we introduce SUN3D, a large-scale RGB-D video database with camera pose and object labels, capturing the full 3D extent of many places. The tasks that go into constructing such a dataset are difficult in isolation -- hand-labeling videos is painstaking, and structure from motion (SfM) is unreliable for large spaces. But if we combine them together, we make the dataset construction task much easier. First, we introduce an intuitive labeling tool that uses a partial reconstruction to propagate labels from one frame to another. Then we use the object labels to fix errors in the reconstruction. For this, we introduce a generalization of bundle adjustment that incorporates object-to-object correspondences. This algorithm works by constraining points for the same object from different frames to lie inside a fixed-size bounding box, parameterized by its rotation and translation. The SUN3D database, the source code for the generalized bundle adjustment, and the web-based 3D annotation tool are all available at http://sun3d.cs.princeton.edu.

We introduce the problem of scene viewpoint recognition, the goal of which is to classify the type of place shown in a photo, and also recognize the observer's viewpoint within that category of place. We construct a database of 360° panoramic images organized into 26 place categories. For each category, our algorithm automatically aligns the panoramas to build a full-view representation of the surrounding place. We also study the symmetry properties and canonical viewpoint of each place category. At test time, given a photo of a scene, the model can recognize the place category, produce a compass-like indication of the observer's most likely viewpoint within that place, and use this information to extrapolate beyond the available view, filling in the probable visual layout that would appear beyond the boundary of the photo.

While long term human visual memory can store a remarkable amount of visual information, it tends to degrade over time. Recent works have shown that image memorability is an intrinsic property of an image that can be reliably estimated using state-of-the-art image features and machine learning algorithms. However, the class of features and image information that is forgotten has not been explored yet. In this work, we propose a probabilistic framework that models how and which local regions from an image may be forgotten using a data-driven approach that combines local and global images features. The model automatically discov- ers memorability maps of individual images without any human annotation. We incorporate multiple image region attributes in our algorithm, leading to improved memorability prediction of images as compared to previous works.

In this paper we seek to detect rectangular cuboids and localize their corners in uncalibrated single-view images depicting everyday scenes. In contrast to recent approaches that rely on detecting vanishing points of the scene and grouping line segments to form cuboids, we build a discriminative parts-based detector that models the appearance of the cuboid corners and internal edges while enforcing consistency to a 3D cuboid model. Our model is invariant to the different 3D viewpoints and aspect ratios and is able to detect cuboids across many different object categories. We introduce a database of images with cuboid annotations that spans a variety of indoor and outdoor scenes and show qualitative and quantitative results on our collected database. Our model out-performs baseline detectors that use 2D constraints alone on the task of localizing cuboid corners.

When glancing at a magazine, or browsing the Internet, we are continuously being exposed to photographs. Despite of this overflow of visual information, humans are extremely good at remembering thousands of pictures along with some of their visual details. But not all images are equal in memory. Some stitch to our minds, and other are forgotten. In this paper we focus on the problem of predicting how memorable an image will be. We show that memorability is a stable property of an image that is shared across different viewers. We introduce a database for which we have measured the probability that each picture will be remembered after a single view. We analyze image features and labels that contribute to making an image memorable, and we train a predictor based on global image descriptors. We find that predicting image memorability is a task that can be addressed with current computer vision techniques. Whereas making memorable images is a challenging task in visualization and photography, this work is a first attempt to quantify this useful quality of images.

Scene categorization is a fundamental problem in computer vision. However, scene understanding research has been constrained by the limited scope of currently-used databases which do not capture the full variety of scene categories. Whereas standard databases for object categorization contain hundreds of different classes of objects, the largest available dataset of scene categories contains only 15 classes. In this paper we propose the extensive Scene UNderstanding (SUN) database that contains 899 categories and 130,519 images. We use 397 well-sampled categories to evaluate numerous state-of-the-art algorithms for scene recognition and establish new bounds of performance. We measure human scene classification performance on the SUN database and compare this with computational methods. Additionally, we study a finer-grained scene representation to detect scenes embedded inside of larger scenes.

We propose an approach that parses registered images captured at ground level into architectural units for large-scale city modeling. Each parsed unit has a regularized shape, which can be used for further modeling purposes. In our approach, we first parse the environment into buildings, the ground, and the sky using a joint 2D-3D segmentation method. Then, we partition buildings into individual façades. The partition problem is formulated as a dynamic programming optimization for a sequence of natural vertical separating lines. Each façade is regularized by a floor line and a roof line. The floor line is the intersection line of the vertical plane of buildings and the horizontal plane of the ground. The roof line links edge points of roof region. The parsed results provide a first geometric approximation to the city environment, and can be further analyzed if necessary. The approach is demonstrated and validated on several large-scale city datasets.

We propose a simple but powerful multi-view semantic segmentation framework for images captured by a camera mounted on a car driving along streets. In our approach, a pair-wise Markov Random Field (MRF) is laid out across multiple views. Both 2D and 3D features are extracted at a super-pixel level to train classifiers for the unary data terms of MRF. For smoothness terms, our approach makes use of color differences in the same image to identify accurate segmentation boundaries, and dense pixel-to-pixel correspondences to enforce consistency across different views. To speed up training and to improve the recognition quality, our approach adaptively selects the most similar training data for each scene from the label pool. Furthermore, we also propose a powerful approach within the same framework to enable large-scale labeling in both the 3D space and 2D images. We demonstrate our approach on more than 10,000 images from Google Maps Street View.

We propose an automatic approach to generate street-side 3D photo-realistic models from images captured along the streets at ground level. We first develop a multi-view semantic segmentation method that recognizes and segments each image at pixel level into semantically meaningful areas, each labeled with a specific object class, such as building, sky, ground, vegetation and car. A partition scheme is then introduced to separate buildings into independent blocks using the major line structures of the scene. Finally, for each block, we propose an inverse patch-based orthographic composition and structure analysis method for façade modeling that efficiently regularizes the noisy and missing reconstructed 3D data. Our system has the distinct advantage of producing visually compelling results by imposing strong priors of building regularity. We demonstrate the fully automatic system on a typical city example to validate our methodology.

In this paper, we introduce a simple sketching method to generate a realistic 3D tree model from a single image. The user draws at least two strokes in the tree image: the first crown stroke around the tree crown to mark up the leaf region, the second branch stroke from the tree root to mark up the main trunk, and possibly few other branch strokes for refinement. The method automatically generates a 3D tree model including branches and leaves. Branches are synthesized by a growth engine from a small library of elementary subtrees that are pre-defined or built on the fly from the recovered visible branches. The visible branches are automatically traced from the drawn branch strokes according to image statistics on the strokes. Leaves are generated from the region bounded by the first crown stroke to complete the tree. We demonstrate our method on a variety of examples.

We propose in this paper a semi-automatic image-based approach to façade modeling that uses images captured along streets and relies on structure from motion to recover camera positions and point clouds automatically as the initial stage for modeling. We start by considering a building façade as a flat rectangular plane or a developable surface with an associated texture image composited from the multiple visible images. A façade is then decomposed and structured into a Directed Acyclic Graph of rectilinear elementary patches. The decomposition is carried out top-down by a recursive subdivision, and followed by a bottom-up merging with the detection of the architectural bilateral symmetry and repetitive patterns. Each subdivided patch of the flat façade is augmented with a depth optimized using the 3D points cloud. Our system also allows for an easy user feedback in the 2D image space for the proposed decomposition and augmentation. Finally, our approach is demonstrated on a large number of façades from a variety of street-side images.

A joint segmentation is a simultaneous segmentation of registered 2D images and 3D points reconstructed from the multiple view images. It is fundamental in structuring the data for subsequent modeling applications. In this paper, we treat this joint segmentation as a weighted graph labeling problem. First, we construct a 3D graph for the joint 3D and 2D points using a joint similarity measure. Then, we propose a hierarchical sparse affinity propagation algorithm to automatically and jointly segment 2D images and group 3D points. Third, a semi-supervised affinity propagation algorithm is proposed to refine the automatic results with the user assistance. Finally, intensive experiments demonstrate the effectiveness of the proposed approaches.