Xuming He

Multiscale conditional random fields for image labeling

We propose an approach to include contextual features for labeling images, in which each pixel is assigned to one of a finite set of labels. The features are incorporated into a probabilistic framework, which combines the outputs of several components. Components differ in the information they encode. Some focus on the image-label mapping, while others focus solely on patterns within the label field. Components also differ in their scale, as some focus on fine-resolution patterns while others on coarser, more global structure. A supervised version of the contrastive divergence algorithm is applied to learn these features from labeled image data. We demonstrate performance on two real-world image databases and compare it to a classifier and a Markov random field.

Learning and incorporating top-down cues in image segmentation

Bottom-up approaches, which rely mainly on continuity principles, are often insufficient to form accurate segments in natural images. In order to improve performance, recent methods have begun to incorporate top-down cues, or object information, into segmentation. In this paper, we propose an approach to utilizing category-based information in segmentation, through a formulation as an image labelling problem. Our approach exploits bottom-up image cues to create an over-segmented representation of an image. The segments are then merged by assigning labels that correspond to the object category. The model is trained on a database of images, and is designed to be modular: it learns a number of image contexts, which simplify training and extend the range of object classes and image database size that the system can handle. The learning method estimates model parameters by maximizing a lower bound of the data likelihood. We examine performance on three real-world image databases, and compare our system to a standard classifier and other conditional random field approaches, as well as a bottom-up segmentation method.

Discrete-Continuous Depth Estimation from a Single Image

In this paper, we tackle the problem of estimating the depth of a scene from a single image. This is a challenging task, since a single image on its own does not provide any depth cue. To address this, we exploit the availability of a pool of images for which the depth is known. More specifically, we formulate monocular depth estimation as a discrete-continuous optimization problem, where the continuous variables encode the depth of the superpixels in the input image, and the discrete ones represent relationships between neighboring superpixels. The solution to this discrete-continuous optimization problem is then obtained by performing inference in a graphical model using particle belief propagation. The unary potentials in this graphical model are computed by making use of the images with known depth. We demonstrate the effectiveness of our model in both the indoor and outdoor scenarios. Our experimental evaluation shows that our depth estimates are more accurate than existing methods on standard datasets.

Multiclass semantic video segmentation with object-level active inference

We address the problem of integrating object reasoning with supervoxel labeling in multiclass semantic video segmentation. To this end, we first propose an object-augmented dense CRF in spatio-temporal domain, which captures long-range dependency between supervoxels, and imposes consistency between object and supervoxel labels. We develop an efficient mean field inference algorithm to jointly infer the supervoxel labels, object activations and their occlusion relations for a moderate number of object hypotheses. To scale up our method, we adopt an active inference strategy to improve the efficiency, which adaptively selects object subgraphs in the object-augmented dense CRF. We formulate the problem as a Markov Decision Process, which learns an approximate optimal policy based on a reward of accuracy improvement and a set of well-designed model and input features. We evaluate our method on three publicly available multiclass video semantic segmentation datasets and demonstrate superior efficiency and accuracy.

Occlusion boundary detection using pseudo-depth

We address the problem of detecting occlusion boundaries from motion sequences, which is important for motion segmentation, estimating depth order, and related tasks. Previous work by Stein and Hebert has addressed this problem and obtained good results on a benchmarked dataset using two-dimensional image cues, motion estimation, and a global boundary model [1]. In this paper we describe a method for detecting occlusion boundaries which uses depth cues and local segmentation cues. More specifically, we show that crude scaled estimates of depth, which we call pseudo-depth, can be extracted from motion sequences containing a small number of image frames using standard SVD factorization methods followed by weak smoothing using a Markov Random Field defined over super-pixels. We then train a classifier for occlusion boundaries using pseudo-depth and local static boundary cues (adding motion cues only gives slightly better results). We evaluate performance on Stein and Heberts dataset and obtain results of similar average quality which are better in the low recall/high precision range. Note that our cues and methods are different from [1] - in particular we did not use their sophisticated global boundary model - and so we conjecture that a unified approach would yield even better results.

Robust Face Alignment Under Occlusion via Regional Predictive Power Estimation

Face alignment has been well studied in recent years, however, when a face alignment model is applied on facial images with heavy partial occlusion, the performance deteriorates significantly. In this paper, instead of training an occlusion-aware model with visibility annotation, we address this issue via a model adaptation scheme that uses the result of a local regression forest (RF) voting method. In the proposed scheme, the consistency of the votes of the local RF in each of several oversegmented regions is used to determine the reliability of predicting the location of the facial landmarks. The latter is what we call regional predictive power (RPP). Subsequently, we adapt a holistic voting method (cascaded pose regression based on random ferns) by putting weights on the votes of each fern according to the RPP of the regions used in the fern tests. The proposed method shows superior performance over existing face alignment models in the most challenging data sets (COFW and 300-W). Moreover, it can also estimate with high accuracy (72.4% overlap ratio) which image areas belong to the face or nonface objects, on the heavily occluded images of the COFW data set, without explicit occlusion modeling.

Superpixel Graph Label Transfer with Learned Distance Metric

We present a fast approximate nearest neighbor algorithm for semantic segmentation. Our algorithm builds a graph over superpixels from an annotated set of training images. Edges in the graph represent approximate nearest neighbors in feature space. At test time we match superpixels from a novel image to the training images by adding the novel image to the graph. A move-making search algorithm allows us to leverage the graph and image structure for finding matches. We then transfer labels from the training images to the image under test. To promote good matches between superpixels we propose to learn a distance metric that weights the edges in our graph. Our approach is evaluated on four standard semantic segmentation datasets and achieves results comparable with the state-of-the-art.

Indoor scene structure analysis for single image depth estimation

We tackle the problem of single image depth estimation, which, without additional knowledge, suffers from many ambiguities. Unlike previous approaches that only reason locally, we propose to exploit the global structure of the scene to estimate its depth. To this end, we introduce a hierarchical representation of the scene, which models local depth jointly with mid-level and global scene structures. We formulate single image depth estimation as inference in a graphical model whose edges let us encode the interactions within and across the different layers of our hierarchy. Our method therefore still produces detailed depth estimates, but also leverages higher-level information about the scene. We demonstrate the benefits of our approach over local depth estimation methods on standard indoor datasets.

An Exemplar-Based CRF for Multi-instance Object Segmentation

We address the problem of joint detection and segmentation of multiple object instances in an image, a key step towards scene understanding. Inspired by data-driven methods, we propose an exemplar-based approach to the task of instance segmentation, in which a set of reference image/shape masks is used to find multiple objects. We design a novel CRF framework that jointly models object appearance, shape deformation, and object occlusion. To tackle the challenging MAP inference problem, we derive an alternating procedure that interleaves object segmentation and shape/appearance adaptation. We evaluate our method on two datasets with instance labels and show promising results.

Learning Structured Hough Voting for Joint Object Detection and Occlusion Reasoning

We propose a structured Hough voting method for detecting objects with heavy occlusion in indoor environments. First, we extend the Hough hypothesis space to include both object location and its visibility pattern, and design a new score function that accumulates votes for object detection and occlusion prediction. In addition, we explore the correlation between objects and their environment, building a depth-encoded object-context model based on RGB-D data. Particularly, we design a layered context representation and allow image patches from both objects and backgrounds voting for the object hypotheses. We demonstrate that using a data-driven 2.1D representation we can learn visual codebooks with better quality, and more interpretable detection results in terms of spatial relationship between objects and viewer. We test our algorithm on two challenging RGB-D datasets with significant occlusion and intraclass variation, and demonstrate the superior performance of our method.