CVPR 2015 Papers

Do object part localization methods produce bilaterally symmetric results on mirror images? Surprisingly not, even though state of the art methods augment the training set with mirrored images. In this paper we take a closer look into this issue. We first introduce the concept of mirrorability as the ability of a model to produce symmetric results in mirrored images and introduce a corresponding measure, namely the \textit{mirror error} that is defined as the difference between the detection result on an image and the mirror of the detection result on its mirror image. We evaluate the mirrorability of several state of the art algorithms in two of the most intensively studied problems, namely human pose estimation and face alignment. Our experiments lead to several interesting findings: 1) Most of state of the art methods struggle to preserve the mirror symmetry, despite the fact that they do have very similar overall performance on the original and mirror images; 2) the low mirrorability is not caused by training or testing sample bias - all algorithms are trained on both the original images and their mirrored versions; 3) the mirror error is strongly correlated to the localization/alignment error (with correlation coefficients around 0.7). Since the mirror error is calculated without knowledge of the ground truth, we show two interesting applications - in the first it is used to guide the selection of difficult samples and in the second to give feedback in a popular Cascaded Pose Regression method for face alignment.

Affordances are fundamental attributes of objects. Affordances reveal the functionalities of objects and the possible actions that can be performed on them. Understanding affordances is crucial for recognizing human activities in visual data and for robots to interact with the world. In this paper we introduce the new problem of mining the knowledge of semantic affordance: given an object, determining whether an action can be performed on it. This is equivalent to connecting verb nodes and noun nodes in WordNet, or filling an affordance matrix encoding the plausibility of each action-object pair. We introduce a new benchmark with crowdsourced ground truth affordances on 20 PASCAL VOC object classes and 957 action classes. We explore a number of approaches including text mining, visual mining, and collaborative filtering. Our analyses yield a number of significant insights that reveal the most effective ways of collecting knowledge of semantic affordances.

In this paper we explore the bi-directional mapping between images and their sentence-based descriptions. Critical to our approach is a recurrent neural network that attempts to dynamically build a visual representation of the scene as a caption is being generated or read. The representation automatically learns to remember long-term visual concepts. Our model is capable of both generating novel captions given an image, and reconstructing visual features given an image description. We evaluate our approach on several tasks. These include sentence generation, sentence retrieval and image retrieval. State-of-the-art results are shown for the task of generating novel image descriptions. When compared to human generated captions, our automatically generated captions are equal to or preferred by humans $21.0\%$ of the time. Results are better than or comparable to state-of-the-art results on the image and sentence retrieval tasks for methods using similar visual features.

Mid-level visual element discovery aims to find clusters of image patches that are both representative and discriminative. In this work, we study this problem from the prospective of pattern mining while relying on the recently popularized Convolutional Neural Networks (CNNs). Specifically, we find that for an image patch, activation extracted from the first fully-connected layer of a CNN have two appealing properties which enable its seamless integration with pattern mining. Patterns are then discovered from a large number of CNN activations of image patches through the well-known association rule mining. When we retrieve and visualize image patches with the same pattern (See Fig. 1), surprisingly, they are not only visually similar but also semantically consistent. We apply our approach to scene and object classification tasks, and demonstrate that our approach outperforms all previous works on mid-level visual element discovery by a sizeable margin with far fewer elements being used. Our approach also outperforms or matches recent works using CNN for these tasks. Source code of the complete system is available online.

Metric learning has proved very successful. However, human annotations are necessary. In this paper, we propose an unsupervised method, dubbed Metric Imitation (MI), where metrics over one cheap feature (target features, TFs) are learned by imitating the standard metrics over another sophisticated, off-the-shelf feature (source features, SFs) by transferring the view-independent property manifold structures. In particular, MI consists of: 1) quantifying the properties of source metrics as manifold geometry, 2) transferring the manifold from source domain to target domain, and 3) learning a mapping of TFs so that the manifold is approximated as well as possible in the mapped feature domain. MI is useful in at least two scenarios where: 1) TFs are more efficient computationally and in terms of memory than SFs; and 2) SFs contain privileged information, but they are not available during testing. For the former, MI is evaluated on image clustering, category-based image retrieval, and instance-based object retrieval, with three SFs and three TFs. For the latter, MI is tested on the task of example-based image super-resolution, where high-resolution patches are taken as SFs and low-resolution patches as TFs. Experiments show that MI is able to provide good metrics while avoiding expensive data labeling efforts and that it achieves state-of-the-art performance for image super-resolution. In addition, manifold transfer is an interesting direction of transfer learning.

Recently, there have been many proposals with state-of-the-art results in subspace clustering that take advantages of the low-rank or sparse optimization techniques. These methods are based on self-expressive models, which have well-defined theoretical aspects. They produce matrices with (approximately) block-diagonal structure, which is then applied to spectral clustering. However, there is no definitive way to construct affinity matrices from these block-diagonal matrices and it is ambiguous how the performance will be affected by the construction method. In this paper, we propose an alternative approach to detect block-diagonal structures from these matrices. The proposed method shares the philosophy of the above subspace clustering methods, in that it is a self-expressive system based on a Hadamard product of a membership matrix. To resolve the difficulty in handling the membership matrix, we solve the convex relaxation of the problem and then transform the representation to a doubly stochastic matrix, which is closely related to spectral clustering. The result of our method has eigenvalues normalized in between zero and one, which is more reliable to estimate the number of clusters and to perform spectral clustering. The proposed method shows competitive results in our experiments, even though we simply count the number of eigenvalues larger than a certain threshold to find the number of clusters.

We consider MAP-inference for graphical models and propose a novel efficient algorithm for finding persistent labels. Our algorithm marks each label in each node of the considered graphical model either as (i) optimal, meaning that it belongs to all optimal solutions of the inference problem; (ii) non-optimal if it provably does not belong to any solution; or (iii) undefined, which means our algorithm can not make a decision regarding the label. Moreover, we prove optimality of our approach, that it delivers in a certain sense the largest total number of labels marked as optimal or non-optimal. We demonstrate superiority of our approach on problems from machine learning and computer vision benchmarks.

In real applications, data is not always explicitly-labeled. For instance, label ambiguity exists when we associate two persons appearing in a news photo with two names provided in the caption. We propose a matrix completion-based method for predicting the actual labels from the ambiguously labeled instances, and a standard supervised classifier can learn from the disambiguated labels to classify new data. We further generalize the method to handle the labeling constraints between instances when such prior knowledge is available. Compared to existing methods, our approach achieves 2.9% improvement on the labeling accuracy of the Lost dataset and comparable performance on the Labeled Yahoo! News dataset.

Recognizing materials in real-world images is a challenging task. Real-world materials have rich surface texture, geometry, lighting conditions, and clutter, which combine to make the problem particularly difficult. In this paper, we introduce a new, large-scale, open dataset of materials in the wild, the Materials in Context Database (MINC), and combine this dataset with deep learning to achieve material recognition and segmentation of images in the wild. MINC is an order of magnitude larger than previous material databases, while being more diverse and well-sampled across its 23 categories. Using MINC, we train convolutional neural networks (CNNs) for two tasks: classifying materials from patches, and simultaneous material recognition and segmentation in full images. For patch-based classification on MINC we found that the best performing CNN architectures can achieve 85.2% mean class accuracy. We convert these trained CNN classifiers into an efficient fully convolutional framework combined with a fully connected conditional random field (CRF) to predict the material at every pixel in an image, achieving 73.1% mean class accuracy. Our experiments demonstrate that having a large, well-sampled dataset such as MINC is crucial for real-world material recognition and segmentation.

Material classification is an important area of research in computer vision. Typical algorithms use color and texture information for classification, but there are problems due to varying lighting conditions and diversity of colors in a single material class. In this work we study the use of long wave infrared (i.e. thermal) imagery for material classification. Thermal imagery has the benefit of relative invariance to color changes, invariance to lighting conditions, and can even work in the dark. We collect a database of 21 different material classes with both color and thermal imagery. We develop a set of features that describe water permeation and heating/cooling properties, and test several variations on these methods to obtain our final classifier. The results show that the proposed method outperforms typical color and texture features, and when combined with color information, the results are improved further.

Both parametric and non-parametric approaches have demonstrated encouraging performances in the human parsing task, namely segmenting a human image into several semantic regions (e.g., hat, bag, left arm, face). In this work, we aim to develop a new solution with the advantages of both methodologies, namely supervision from annotated data and the flexibility to use newly annotated (possibly uncommon) images, and present a quasi-parametric human parsing model. Under the classic KNN-based nonparametric framework, the parametric Matching Convolutional Neural Network (M-CNN) is proposed to predict the matching confidence and displacement of the best matched region in the testing image for a particular semantic region in one KNN image. Given a testing image, we first retrieve its KNN images from the annotated/manually-parsed human image corpus. Then each semantic region in each KNN image is matched with confidence to the testing image using M-CNN, and the matched regions from all KNN images are further fused, followed by a superpixel smoothing procedure to obtain the ultimate human parsing result. The M-CNN differs from the classic CNN in that the tailored cross image matching filters are introduced to characterize the matching between the testing image and the semantic region of a KNN image. The cross image matching filters are defined at different convolution layers, each aiming to capture a particular range of displacements. Comprehensive evaluations over a large dataset with 7,700 annotated human images well demonstrate the significant performance gain from the quasi-parametric model over the state-of-the-arts, for the human parsing task.

We study the new problem of matching regions between a pair of RGBD images given a large set of overlapping region proposals. These region proposals do not have a tree hierarchy and are treated as bags of regions. Matching RGBD images using bags of region candidates with unstructured relations is a challenging combinatorial problem. We propose a linear formulation, which optimizes the region selection and matching simultaneously so that the matched regions have similar color histogram, shape, and small overlaps, the selected regions have a small number and overall low concavity, and they tend to cover both of the images. We efficiently compute the lower bound by solving a sequence of min-cost bipartite matching problems via Lagrangian relaxation and we obtain the global optimum using branch and bound. Our experiments show that the proposed method is fast, accurate, and robust against cluttered scenes.

Motivated by recent successes on learning feature representations and on learning feature comparison functions, we propose a unified approach to combining both for training a patch matching system. Our system, dubbed MatchNet, consists of a deep convolutional network that extracts features from patches and a network of three fully connected layers that computes a similarity between the extracted features. To ensure experimental repeatability, we train MatchNet on standard datasets and employ an input sampler to augment the training set with synthetic exemplar pairs that reduce overfitting. Once trained, we achieve better computational efficiency during matching by disassembling MatchNet and separately applying the feature computation and similarity networks in two sequential stages. We perform a comprehensive set of experiments on standard datasets to carefully study the contributions of each aspect of MatchNet, with direct comparisons to established methods. Our results confirm that our unified approach improves accuracy over previous state-of-the-art results on patch matching datasets, while reducing the storage requirement for descriptors. We make pre-trained MatchNet publicly available.

Content based image retrieval is highly relevant in medical imaging, since it makes vast amounts of imaging data accessible for comparison during diagnosis. Finding image similarity measures that reflect diagnostically relevant relationships is challenging, since the overall appearance variability is high compared to often subtle signatures of diseases. To learn models that capture the relationship between semantic clinical information and image elements at scale, we have to rely on data generated during clinical routine (images and radiology reports), since expert annotation is prohibitively costly. Here we show that re-mapping visual features extracted from medical imaging data based on weak labels that can be found in corresponding radiology reports creates descriptions of local image content capturing clinically relevant information. We show that these semantic profiles enable higher recall and precision during retrieval compared to visual features, and that we can even map semantic terms describing clinical findings from radiology reports to localized image volume areas.

We propose a perceptual grouping framework that organizes image edges into meaningful structures and demonstrate its usefulness on various computer vision tasks. Our grouper formulates edge grouping as a graph partition problem, where a learning to rank method is developed to encode probabilities of candidate edge pairs. In particular, RankSVM is employed for the first time to combine multiple Gestalt principles as cue for edge grouping. Afterwards, an edge grouping based object proposal measure is introduced that yields proposals comparable to state-of-the-art alternatives. We further show how human-like sketches can be generated from edge groupings and consequently used to deliver state-of-the-art sketch-based image retrieval performance. Last but not least, we tackle the problem of freehand human sketch segmentation by utilizing the proposed grouper to cluster strokes into semantic object parts.

Variations in the appearance of a tracked object, such as changes in geometry/photometry, camera viewpoint, illumination, or partial occlusion, pose a major challenge to object tracking. Here, we adopt cognitive psychology principles to design a flexible representation that can adapt to changes in object appearance during tracking. Inspired by the well-known Atkinson-Shiffrin Memory Model, we propose MUlti-Store Tracker (MUSTer), a dual-component approach consisting of short- and long-term memory stores to process target appearance memories. A powerful and efficient Integrated Correlation Filter (ICF) is employed in the short-term store for short-term tracking. The integrated long-term component, which is based on keypoint matching-tracking and RANSAC estimation, can interact with the long-term memory and provide additional information for output control. MUSTer was extensively evaluated on the CVPR2013 Online Object Tracking Benchmark (OOTB) and ALOV++ datasets. The experimental results demonstrated the superior performance of MUSTer in comparison with other state-of-art trackers.

Graph cuts-based algorithms have achieved great success in energy minimization for many computer vision applications. These algorithms provide approximated solutions for multi-label energy functions via move-making approach. This approach fuses the current solution with a proposal to generate a lower-energy solution. Thus, generating the appropriate proposals is necessary for the success of the move-making approach. However, not much research efforts has been done on the generation of ``good'' proposals, especially for non-metric energy functions. In this paper, we propose an application-independent and energy-based approach to generate ``good'' proposals. With these proposals, we present a graph cuts-based move-making algorithm called GA-fusion (fusion with graph approximation-based proposals). Extensive experiments support that our proposal generation is effective across different classes of energy functions. The proposed algorithm outperforms others both on real and synthetic problems.

We introduce a multi-scale framework for low-level vision, where the goal is estimating physical scene values from image data---such as depth from stereo image pairs. The framework uses a dense, overlapping set of image regions at multiple scales and a ``local model,'' such as a slanted-plane model for stereo disparity, that is expected to be valid piecewise across the visual field. Estimation is cast as optimization over a dichotomous mixture of variables, simultaneously determining which regions are inliers with respect to the local model (binary variables) and the correct co-ordinates in the local model space for each inlying region (continuous variables). When the regions are organized into a multi-scale hierarchy, optimization can occur in an efficient and parallel architecture, where distributed computational units iteratively perform calculations and share information through sparse connections between parents and children. The framework performs well on a standard benchmark for binocular stereo, and it produces a distributional scene representation that is appropriate for combining with higher-level reasoning and other low-level cues.

Models comprised of deep convolutional network layers have dominated recent image interpretation tasks; we investigate whether models which are also compositional, or "deep", temporally are effective on tasks involving visual sequences or label sequences. We develop a novel recurrent convolutional architecture suitable for large-scale visual learning which is end-to-end trainable, and demonstrate the value of these models on benchmark video recognition tasks, image to sentence generation problems, and video narration challenges. In contrast to current models which assume a fixed spatio-temporal receptive field or simple temporal averaging for sequential processing, recurrent convolutional models are "doubly deep" in that they can be compositional in spatial and temporal "layers". Such models may have advantages when target concepts are complex and/or training data are limited. Learning long-term dependencies is possible when nonlinearities are incorporated into the network state updates. Long-term RNN models are appealing in that they directly can map variable length inputs (i.e. video frames) to variable length outputs (i.e. natural language text) and can model complex temporal dynamics; yet they can be optimized with backpropagation. Our recurrent long-term models are directly connected to state-of-the-art visual convnet models and can jointly trained, updating temporal dynamics and convolutional perceptual representations simultaneously. Our results show such models have distinct advantages over state-of-the-art models for recognition or generation which are separately defined and/or optimized.