Search

Hashing is a popular search algorithm for its compact binary representation and efficient Hamming distance calculation. Benefited from the advance of deep learning, deep hashing methods have achieved promising performance. However, those methods usually learn with expensive labeled data but fail to utilize unlabeled data. Furthermore, the traditional pairwise loss used by those methods cannot explicitly force similar/dissimilar pairs to small/large distances. Both weaknesses limit existing methods’ performance. To solve the first problem, we propose a novel semi-supervised deep hashing model named adversarial binary mutual learning (ABML). Specifically, our ABML consists of a generative model $G_{H}$ and a discriminative model $D_{H}$ , where $D_{H}$ learns labeled data in a supervised way and $G_{H}$ learns unlabeled data by synthesizing real images. We adopt an adversarial learning (AL) strategy to transfer the knowledge of unlabeled data to $D_{H}$ by making $G_{H}$ and $D_{H}$ mutually learn from each other. To solve the second problem, we propose a novel Weibull cross-entropy loss (WCE) by using the Weibull distribution, which can distinguish tiny differences of distances and explicitly force similar/dissimilar distances as small/large as possible. Thus, the learned features are more discriminative. Finally, by incorporating ABML with WCE loss, our model can acquire more semantic and discriminative features. Extensive experiments on four common data sets (CIFAR-10, large database of handwritten digits (MNIST), ImageNet-10, and NUS-WIDE) and a large-scale data set ImageNet demonstrate that our approach successfully overcomes the two difficulties above and significantly outperforms state-of-the-art hashing methods. [ABSTRACT FROM AUTHOR]

The deep neural network (DNN) has achieved remarkable performance in a wide range of applications at the cost of huge memory and computational complexity. Fixed-point network quantization emerges as a popular acceleration and compression method but still suffers from huge performance degradation when extremely low-bit quantization is utilized. Moreover, current fixed-point quantization methods rely heavily on supervised retraining using large amounts of the labeled training data, while the labeled data are hard to obtain in the real-world applications. In this article, we propose an efficient framework, namely, fixed-point factorized network (FFN), to turn all weights into ternary values, i.e., {−1, 0, 1}. We highlight that the proposed FFN framework can achieve negligible degradation even without any supervised retraining on the labeled data. Note that the activations can be easily quantized into an 8-bit format; thus, the resulting networks only have low-bit fixed-point additions that are significantly more efficient than 32-bit floating-point multiply–accumulate operations (MACs). Extensive experiments on large-scale ImageNet classification and object detection on MS COCO show that the proposed FFN can achieve about more than 20 × compression and remove most of the multiply operations with comparable accuracy. Codes are available on GitHub at https://github.com/wps712/FFN. [ABSTRACT FROM AUTHOR]

Object-centric-based categorization methods have been proven more effective than hard partitions of images (e.g., spatial pyramid matching). However, how to determine the locations of objects is still an open problem. Besides, modeling of context areas is often mixed with the background. Moreover, the semantic information is often ignored by these methods that only use visual representations for classification. In this paper, we propose an object categorization method by semantically modeling the object and context information (SOC). We first select a number of candidate regions with high confidence scores and semantically represent these regions by measuring correlations of each region with prelearned classifiers (e.g., local feature-based classifiers and deep convolutional-neural-network-based classifiers). These regions are clustered for object selections. The other selected areas are then viewed as context areas. We treat other areas beyond the object and context areas within one image as the background. The visually and semantically represented objects and contexts are then used along with the background area for object representations and categorizations. Experimental results on several public data sets well demonstrate the effectiveness of the proposed object categorization method by semantically modeling the object and context information. [ABSTRACT FROM AUTHOR]

We are witnessing an explosive development and widespread application of deep neural networks (DNNs) in various fields. However, DNN models, especially a convolutional neural network (CNN), usually involve massive parameters and are computationally expensive, making them extremely dependent on high-performance hardware. This prohibits their further extensions, e.g., applications on mobile devices. In this paper, we present a quantized CNN, a unified approach to accelerate and compress convolutional networks. Guided by minimizing the approximation error of individual layer’s response, both fully connected and convolutional layers are carefully quantized. The inference computation can be effectively carried out on the quantized network, with much lower memory and storage consumption. Quantitative evaluation on two publicly available benchmarks demonstrates the promising performance of our approach: with comparable classification accuracy, it achieves 4 to $6 \times $ acceleration and 15 to $20\times $ compression. With our method, accurate image classification can even be directly carried out on mobile devices within 1 s. [ABSTRACT FROM AUTHOR]

Object categorization refers to the task of automatically classifying objects based on the visual content. Existing approaches simply represent each image with the visual features without considering the specific characters of images within the same class. However, objects of the same class may exhibit unique characters, which should be represented accordingly. In this brief, we propose a novel class-specific representation strategy for object categorization. For each class, we first model the characters of images within the same class using Gaussian mixture model (GMM). We then represent each image by calculating the Euclidean distance and relative Euclidean distance between the image and the GMM model for each class. We concatenate the representations of each class for joint representation. In this way, we can represent an image by not only considering the visual contents but also combining the class-specific characters. Experiments on several public available data sets validate the superiority of the proposed class-specific representation method over well-established algorithms for object category predictions. [ABSTRACT FROM AUTHOR]

Most image classification methods try to learn classifiers for each class using training images alone. Due to the interclass and intraclass variations, it would be more effective to take the testing images into consideration for classifier learning. In this brief, we propose a novel image-specific classification method by combing the local and global discriminations of training images. We adaptively train classifier for each testing image instead of generating classifiers for each class with training images alone. For each testing image, we first select its ${k}$ nearest neighbors in the training set with the corresponding labels for local classifier training. This helps to model the distinctive characters of each testing image. Besides, we also use all the training images for global discrimination modeling. The local and global discriminations are combined for final classification. In this way, we could not only model the specific character of each testing image but also avoid the local optimum by jointly considering all the training images. To evaluate the usefulness of the proposed image-specific classification with local and global discrimination (ISC-LG) method, we conduct image classification experiments on several public image data sets. The superior performances over other baseline methods prove the effectiveness of the proposed ISC-LG method. [ABSTRACT FROM AUTHOR]

Visual features have been widely used for image representation and categorization. However, visual features are often inconsistent with human perception. Besides, constructing explicit semantic space is still an open problem. To alleviate these two problems, in this paper, we propose to construct structured weak semantic space for image representation. Exemplar classifier is first trained to separate each training image from other images for weak semantic space construction. However, each exemplar classifier separates one training image from other images, and it only has limited semantic separability. Besides, the outputs of exemplar classifiers are inconsistent with each other. We jointly construct the weak semantic space using structured constraint. This is achieved by imposing low-rank constraint on the outputs of exemplar classifiers with sparsity constraint. An alternative optimization procedure is used to learn the exemplar classifiers. Since the proposed method does not dependent on the initial image representation strategy, we can make use of various visual features for efficient exemplar classifier training (e.g., fisher vector-based methods and convolutional neural networks-based methods). We apply the proposed structured weak semantic space-based image representation method for categorization. The experimental results on several public image data sets prove the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]