When a speaker says the name of a color, the color that they picture is not necessarily the same as the listener imagines. Color is a grounded semantic task, but that grounding is not a mapping of a single word (or phrase) to a single point in color-space. Proper understanding of color language requires the capacity to map a sequence of words to a probability distribution in color-space. A distribution is required as there is no clear agreement between people as to what a particular color describes -- different people have a different idea of what it means to be `very dark orange'. We propose a novel GRU-based model to handle this case. Learning how each word in a color name contributes to the color described, allows for knowledge sharing between uses of the words in different color names. This knowledge sharing significantly improves predicative capacity for color names with sparse training data. The extreme case of this challenge in data sparsity is for color names without any direct training data. Our model is able to predict reasonable distributions for these cases, as evaluated on a held-out dataset consisting only of such terms.
This paper proposes a novel image set classification technique based on the concept of linear regression. Unlike most other approaches, the proposed technique does not involve any training or feature extraction. The gallery image sets are represented as subspaces in a high dimensional space. Class specific gallery subspaces are used to estimate regression models for each image of the test image set. Images of the test set are then projected on the gallery subspaces. Residuals, calculated using the Euclidean distance between the original and the projected test images, are used as the distance metric. Three different strategies are devised to decide on the final class of the test image set. We performed extensive evaluations of the proposed technique under the challenges of low resolution, noise and less gallery data for the tasks of surveillance, video based face recognition and object recognition. Experiments show that the proposed technique achieves a better classification accuracy and a faster execution time under the challenging testing conditions.
By introducing sign constraints on the weights, this paper proposes sign constrained rectifier networks (SCRNs), whose training can be solved efficiently by the well known majorization-minimization (MM) algorithms. We prove that the proposed two-hidden-layer SCRNs, which exhibit negative weights in the second hidden layer and negative weights in the output layer, are capable of separating any two (or more) disjoint pattern sets. Furthermore, the proposed two-hidden-layer SCRNs can decompose the patterns of each class into several clusters so that each cluster is convexly separable from all the patterns from the other classes. This provides a means to learn the pattern structures and analyse the discriminant factors between different classes of patterns.
This paper provides a theoretical justification of the superior classification performance of deep rectifier networks over shallow rectifier networks from the geometrical perspective of piecewise linear (PWL) classifier boundaries. We show that, for a given threshold on the approximation error, the required number of boundary facets to approximate a general smooth boundary grows exponentially with the dimension of the data, and thus the number of boundary facets, referred to as boundary resolution, of a PWL classifier is an important quality measure that can be used to estimate a lower bound on the classification errors. However, learning naively an exponentially large number of boundary facets requires the determination of an exponentially large number of parameters and also requires an exponentially large number of training patterns. To overcome this issue of "curse of dimensionality", compressive representations of high resolution classifier boundaries are required. To show the superior compressive power of deep rectifier networks over shallow rectifier networks, we prove that the maximum boundary resolution of a single hidden layer rectifier network classifier grows exponentially with the number of units when this number is smaller than the dimension of the patterns. When the number of units is larger than the dimension of the patterns, the growth rate is reduced to a polynomial order. Consequently, the capacity of generating a high resolution boundary will increase if the same large number of units are arranged in multiple layers instead of a single hidden layer. Taking high dimensional spherical boundaries as examples, we show how deep rectifier networks can utilize geometric symmetries to approximate a boundary with the same accuracy but with a significantly fewer number of parameters than single hidden layer nets.
Predicting an interaction before it is fully executed is very important in applications such as human-robot interaction and video surveillance. In a two-human interaction scenario, there often contextual dependency structure between the global interaction context of the two humans and the local context of the different body parts of each human. In this paper, we propose to learn the structure of the interaction contexts, and combine it with the spatial and temporal information of a video sequence for a better prediction of the interaction class. The structural models, including the spatial and the temporal models, are learned with Long Short Term Memory (LSTM) networks to capture the dependency of the global and local contexts of each RGB frame and each optical flow image, respectively. LSTM networks are also capable of detecting the key information from the global and local interaction contexts. Moreover, to effectively combine the structural models with the spatial and temporal models for interaction prediction, a ranking score fusion method is also introduced to automatically compute the optimal weight of each model for score fusion. Experimental results on the BIT Interaction and the UT-Interaction datasets clearly demonstrate the benefits of the proposed method.
This paper presents a new method for 3D action recognition with skeleton sequences (i.e., 3D trajectories of human skeleton joints). The proposed method first transforms each skeleton sequence into three clips each consisting of several frames for spatial temporal feature learning using deep neural networks. Each clip is generated from one channel of the cylindrical coordinates of the skeleton sequence. Each frame of the generated clips represents the temporal information of the entire skeleton sequence, and incorporates one particular spatial relationship between the joints. The entire clips include multiple frames with different spatial relationships, which provide useful spatial structural information of the human skeleton. We propose to use deep convolutional neural networks to learn long-term temporal information of the skeleton sequence from the frames of the generated clips, and then use a Multi-Task Learning Network (MTLN) to jointly process all frames of the generated clips in parallel to incorporate spatial structural information for action recognition. Experimental results clearly show the effectiveness of the proposed new representation and feature learning method for 3D action recognition.
Conventional change detection methods require a large number of images to learn background models or depend on tedious pixel-level labeling by humans. In this paper, we present a weakly supervised approach that needs only image-level labels to simultaneously detect and localize changes in a pair of images. To this end, we employ a deep neural network with DAG topology to learn patterns of change from image-level labeled training data. On top of the initial CNN activations, we define a CRF model to incorporate the local differences and context with the dense connections between individual pixels. We apply a constrained mean-field algorithm to estimate the pixel-level labels, and use the estimated labels to update the parameters of the CNN in an iterative EM framework. This enables imposing global constraints on the observed foreground probability mass function. Our evaluations on four benchmark datasets demonstrate superior detection and localization performance.
In this paper, we introduce transformations of deep rectifier networks, enabling the conversion of deep rectifier networks into shallow rectifier networks. We subsequently prove that any rectifier net of any depth can be represented by a maximum of a number of functions that can be realized by a shallow network with a single hidden layer. The transformations of both deep rectifier nets and deep residual nets are conducted to demonstrate the advantages of the residual nets over the conventional neural nets and the advantages of the deep neural nets over the shallow neural nets. In summary, for two rectifier nets with different depths but with same total number of hidden units, the corresponding single hidden layer representation of the deeper net is much more complex than the corresponding single hidden representation of the shallower net. Similarly, for a residual net and a conventional rectifier net with the same structure except for the skip connections in the residual net, the corresponding single hidden layer representation of the residual net is much more complex than the corresponding single hidden layer representation of the conventional net.
Class imbalance is a common problem in the case of real-world object detection and classification tasks. Data of some classes is abundant making them an over-represented majority, and data of other classes is scarce, making them an under-represented minority. This imbalance makes it challenging for a classifier to appropriately learn the discriminating boundaries of the majority and minority classes. In this work, we propose a cost sensitive deep neural network which can automatically learn robust feature representations for both the majority and minority classes. During training, our learning procedure jointly optimizes the class dependent costs and the neural network parameters. The proposed approach is applicable to both binary and multi-class problems without any modification. Moreover, as opposed to data level approaches, we do not alter the original data distribution which results in a lower computational cost during the training process. We report the results of our experiments on six major image classification datasets and show that the proposed approach significantly outperforms the baseline algorithms. Comparisons with popular data sampling techniques and cost sensitive classifiers demonstrate the superior performance of our proposed method.
We propose a novel image set classification technique using linear regression models. Downsampled gallery image sets are interpreted as subspaces of a high dimensional space to avoid the computationally expensive training step. We estimate regression models for each test image using the class specific gallery subspaces. Images of the test set are then reconstructed using the regression models. Based on the minimum reconstruction error between the reconstructed and the original images, a weighted voting strategy is used to classify the test set. We performed extensive evaluation on the benchmark UCSD/Honda, CMU Mobo and YouTube Celebrity datasets for face classification, and ETH-80 dataset for object classification. The results demonstrate that by using only a small amount of training data, our technique achieved competitive classification accuracy and superior computational speed compared with the state-of-the-art methods.