Search

Deep neural networks have achieved outstanding performance in computer vision tasks. Convolutional Neural Networks (CNNs) typically operate in the spatial domain with raw images, but in practice, images are usually stored and transmitted in their compressed representation where JPEG is one of the most widely used encoder. Also, these networks are computationally intensive and slow. This paper proposes performing the learning and inference processes in the compressed domain in order to reduce the computational complexity and improve the speed of popular CNNs. For this purpose, a novel graph-based frequency channel selection method is proposed to identify and select the most important frequency channels. The computational complexity is reduced by retaining the important frequency components and discarding the insignificant ones as well as eliminating the unnecessary layers of the network. Experimental results show that the modified ResNet-50 operating in the compressed domain is up to 70% faster than the spatial-based traditional ResNet-50 while resulting in similar classification accuracy. Moreover, this paper proposes a preprocessing step with partial encoding to improve the resilience to distortions caused by low-quality encoded images. Finally, we show that training a network with highly compressed data can achieve a good classification accuracy with up to 93% reduction in the storage requirements of the training data. [ABSTRACT FROM AUTHOR]

Deep neural networks have achieved great success in solving many machine learning and computer vision problems. In this paper, we propose a deep neural network called the Tucker network derived from the Tucker format and analyze its expressive power. The results demonstrate that the Tucker network has exponentially higher expressive power than the shallow network. In other words, a shallow network with an exponential width is required to realize the same score function as that computed by the Tucker network. Moreover, we discuss the expressive power between the hierarchical Tucker tensor network (HT network) and the proposed Tucker network. To generalize the Tucker network into a deep version, we combine the hierarchical Tucker format and Tucker format to propose a deep Tucker tensor decomposition. Its corresponding deep Tucker network is presented. Experiments are conducted on three datasets: MNIST, CIFAR-10 and CIFAR-100. The results experimentally validate the theoretical results and show that the Tucker network and deep Tucker network have better performance than the shallow network and HT network. [ABSTRACT FROM AUTHOR]

Although great success has been achieved in various computer vision tasks, deep neural networks (DNNs) suffer dramatic performance degradation when evaluated on out-of-distribution data. Domain generalization (DG) is proposed to handle this problem by learning domain-agnostic information from multiple source domains to generalize well on unseen target domains. Several methods resort to Fourier transform due to its simplicity and efficiency. They argue that amplitude spectra imply domain-specific information, which should be suppressed, while phase counterparts imply domain-agnostic information, which should be preserved. However, these methods only suppress the domain-specific information in source domains and neglect the relationship with target domains, leading to the persistence of the domain gap. Besides, these methods preserve domain-agnostic information by keeping phase components unchanged, causing them to be underutilized. In this paper, we propose Dual Branch Augmentation Module (DBAM) by leveraging Fourier transform and taking advantage of both amplitude and phase spectra. For the amplitude branch, we propose Inner-domain Amplitude Distribution Rectification (IADR) and Cross-domain Amplitude Dirichlet Mixup (CADM) to stabilize the training process and explore more feature space. In addition, we propose Test-time Amplitude Prototype Calibration (TAPC) to construct the connection between source and target domains during evaluation to further mitigate the domain gap. For the phase branch, we propose Random Symmetric Phase Perturbation (RSPP) to enhance the robustness for recognizing domain-agnostic information. With the combined contributions of the two branches, DBAM significantly surpasses other state-of-the-art (SOTA) methods. Extensive experiments on four benchmarks and further analysis demonstrate the effectiveness of DBAM. [ABSTRACT FROM AUTHOR]

Many large and complex deep neural networks have been shown to provide higher performance on various computer vision tasks. However, very little is known about the relationship between the complexity of the input data along with the type of noise and the depth needed for correct classification. Existing studies do not address the issue of common corruptions adequately, especially in understanding what impact these corruptions leave on the individual part of a deep neural network. Therefore, we can safely assume that the classification (or misclassification) might be happening at a particular layer(s) of a network that accumulates to draw a final correct or incorrect prediction. In this paper, we introduce a novel concept of corruption depth , which identifies the location of the network layer/depth until the misclassification persists. We assert that the identification of such layers will help in better designing the network by pruning certain layers in comparison to the purification of the entire network which is computationally heavy. Through our extensive experiments, we present a coherent study to understand the processing of examples through the network. Our approach also illustrates different philosophies of example memorization and a one-dimensional view of sample or query difficulty. We believe that the understanding of the corruption depth can open a new dimension of model explainability and model compression , where in place of just visualizing the attention map, the classification progress can be seen throughout the network. • A novel metric corruption depth is proposed to explain functioning of neural network. • The proposed research helps in deciphering the flow of classification decisions. • The proposed research overcomes the limitations of understanding example difficulty. • Corruption depth notation helps in pruning the networks while improving the accuracy. [ABSTRACT FROM AUTHOR]

Salient object detection can be applied as a critical preprocessing step in many computer vision tasks. Recent studies of salient object detection mainly employed convolutional neural networks (CNNs) for mining high-level semantic properties. However, the existing methods can still be improved to find precise semantic information in different scenarios. In particular, in the two main methods employed for salient object detection, the patchwise detection models might ignore spatial structures among regions and the fully convolution-based models mainly consider semantic features in a global manner. In this paper, we proposed a salient object detection framework by embedding topological features into a deep neural network for extracting semantics. We segment the input image and compute weight for each region with low-level features. The weighted segmentation result is called a topological map and it provides an additional channel for the CNN to emphasize the structural integrity and locality during the extraction of semantic features. We also utilize the topological map for saliency refinement based on a conditional random field at the end of our model. Experimental results on six benchmark datasets demonstrated that our proposed framework achieves competitive performance compared to other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Over the past decade, Deep Convolutional Neural Networks (DCNNs) have shown remarkable performance in most computer vision tasks. These tasks traditionally use a fixed dataset, and the model, once trained, is deployed as is. Adding new information to such a model presents a challenge due to complex training issues, such as "catastrophic forgetting", and sensitivity to hyper-parameter tuning. However, in this modern world, data is constantly evolving, and our deep learning models are required to adapt to these changes. In this paper, we propose an adaptive hierarchical network structure composed of DCNNs that can grow and learn as new data becomes available. The network grows in a tree-like fashion to accommodate new classes of data, while preserving the ability to distinguish the previously trained classes. The network organizes the incrementally available data into feature-driven super-classes and improves upon existing hierarchical CNN models by adding the capability of self-growth. The proposed hierarchical model, when compared against fine-tuning a deep network, achieves significant reduction of training effort, while maintaining competitive accuracy on CIFAR-10 and CIFAR-100. [ABSTRACT FROM AUTHOR]

Abstract: The Pareto-optimality concept is used in this paper in order to represent a constrained set of solutions that are able to trade-off the two main objective functions involved in neural networks supervised learning: data-set error and network complexity. The neural network is described as a dynamic system having error and complexity as its state variables and learning is presented as a process of controlling a learning trajectory in the resulting state space. In order to control the trajectories, sliding mode dynamics is imposed to the network. It is shown that arbitrary learning trajectories can be achieved by maintaining the sliding mode gains within their convergence intervals. Formal proofs of convergence conditions are therefore presented. The concept of trajectory learning presented in this paper goes further beyond the selection of a final state in the Pareto set, since it can be reached through different trajectories and states in the trajectory can be assessed individually against an additional objective function. [Copyright &y& Elsevier]

Abstract: The neocognitron, which was proposed by , is a hierarchical multi-layered neural network capable of robust visual pattern recognition. It acquires the ability to recognize patterns through learning. This paper proposes a new rule for competitive learning, named winner-kill-loser, and apply it to the neocognitron. The winner-kill-loser rule resembles the winner-take-all rule. Every time when a training stimulus is presented, non-silent cells compete with each other. The winner, however, not only takes all, but also kills losers. In other words, the winner learns the training stimulus, and losers are removed from the network. If all cells are silent, a new cell is generated and it learns the training stimulus. Thus feature-extracting cells gradually come to distribute uniformly in the feature space. The use of winner-kill-loser rule is not limited to the neocognitron. It is useful for various types of competitive learning, in general. This paper also proposes several improvements made on the neocognitron: such as, disinhibition to the inhibitory surround in the connections to C-cells (or complex cells) from S-cells (or simple cells); and square root shaped saturation in the input-to-output characteristics of C-cells. As a result of these improvements, the recognition rate of the neocognitron has been largely increased. [Copyright &y& Elsevier]

Abstract: In this paper, existence and uniqueness of the solution of interval fuzzy Cohen–Grossberg neural networks with piecewise constant argument are discussed. Based on the comparison principle, it presents new theoretical results on the global robust exponential stability of interval fuzzy Cohen–Grossberg networks with piecewise constant argument. As a special case, the corresponding results of interval fuzzy recurrent neural networks with piecewise constant argument are derived. Three examples are given for illustrating validity of the obtained results. [Copyright &y& Elsevier]

Abstract: Object selection refers to the mechanism of extracting objects of interest while ignoring other objects and background in a given visual scene. It is a fundamental issue for many computer vision and image analysis techniques and it is still a challenging task to artificial visual systems. Chaotic phase synchronization takes place in cases involving almost identical dynamical systems and it means that the phase difference between the systems is kept bounded over the time, while their amplitudes remain chaotic and may be uncorrelated. Instead of complete synchronization, phase synchronization is believed to be a mechanism for neural integration in brain. In this paper, an object selection model is proposed. Oscillators in the network representing the salient object in a given scene are phase synchronized, while no phase synchronization occurs for background objects. In this way, the salient object can be extracted. In this model, a shift mechanism is also introduced to change attention from one object to another. Computer simulations show that the model produces some results similar to those observed in natural vision systems. [Copyright &y& Elsevier]

Abstract: In this paper, we propose the concept of a manifold of color perception through empirical observation that the center-surround properties of images in a perceptually similar environment define a manifold in the high dimensional space. Such a manifold representation can be learned using a novel recurrent neural network based learning algorithm. Unlike the conventional recurrent neural network model in which the memory is stored in an attractive fixed point at discrete locations in the state space, the dynamics of the proposed learning algorithm represent memory as a nonlinear line of attraction. The region of convergence around the nonlinear line is defined by the statistical characteristics of the training data. This learned manifold can then be used as a basis for color correction of the images having different color perception to the learned color perception. Experimental results show that the proposed recurrent neural network learning algorithm is capable of color balance the lighting variations in images captured in different environments successfully. [Copyright &y& Elsevier]

Abstract: Over successive stages, the visual system develops neurons that respond with view, size and position invariance to objects or faces. A number of computational models have been developed to explain how transform-invariant cells could develop in the visual system. However, a major limitation of computer modelling studies to date has been that the visual stimuli are typically presented one at a time to the network during training. In this paper, we investigate how vision models may self-organize when multiple stimuli are presented together within each visual image during training. We show that as the number of independent stimuli grows large enough, standard competitive neural networks can suddenly switch from learning representations of the multi-stimulus input patterns to representing the individual stimuli. Furthermore, the competitive networks can learn transform (e.g. position or view) invariant representations of the individual stimuli if the network is presented with input patterns containing multiple transforming stimuli during training. Finally, we extend these results to a multi-layer hierarchical network model (VisNet) of the ventral visual system. The network is trained on input images containing multiple rotating 3D objects. We show that the network is able to develop view-invariant representations of the individual objects. [Copyright &y& Elsevier]