-
RetinotopicNet: An Iterative Attention Mechanism Using Local Descriptors with Global Context
Authors:
Thomas Kurbiel,
Shahrzad Khaleghian
Abstract:
Convolutional Neural Networks (CNNs) were the driving force behind many advancements in Computer Vision research in recent years. This progress has spawned many practical applications and we see an increased need to efficiently move CNNs to embedded systems today. However traditional CNNs lack the property of scale and rotation invariance: two of the most frequently encountered transformations in…
▽ More
Convolutional Neural Networks (CNNs) were the driving force behind many advancements in Computer Vision research in recent years. This progress has spawned many practical applications and we see an increased need to efficiently move CNNs to embedded systems today. However traditional CNNs lack the property of scale and rotation invariance: two of the most frequently encountered transformations in natural images. As a consequence CNNs have to learn different features for same objects at different scales. This redundancy is the main reason why CNNs need to be very deep in order to achieve the desired accuracy. In this paper we develop an efficient solution by reproducing how nature has solved the problem in the human brain. To this end we let our CNN operate on small patches extracted using the log-polar transform, which is known to be scale and rotation equivariant. Patches extracted in this way have the nice property of magnifying the central field and compressing the periphery. Hence we obtain local descriptors with global context information. However the processing of a single patch is usually not sufficient to achieve high accuracies in e.g. classification tasks. We therefore successively jump to several different locations, called saccades, thus building an understanding of the whole image. Since log-polar patches contain global context information, we can efficiently calculate following saccades using only the small patches. Saccades efficiently compensate for the lack of translation equivariance of the log-polar transform.
△ Less
Submitted 12 May, 2020;
originally announced May 2020.
-
Training of Deep Neural Networks based on Distance Measures using RMSProp
Authors:
Thomas Kurbiel,
Shahrzad Khaleghian
Abstract:
The vanishing gradient problem was a major obstacle for the success of deep learning. In recent years it was gradually alleviated through multiple different techniques. However the problem was not really overcome in a fundamental way, since it is inherent to neural networks with activation functions based on dot products. In a series of papers, we are going to analyze alternative neural network st…
▽ More
The vanishing gradient problem was a major obstacle for the success of deep learning. In recent years it was gradually alleviated through multiple different techniques. However the problem was not really overcome in a fundamental way, since it is inherent to neural networks with activation functions based on dot products. In a series of papers, we are going to analyze alternative neural network structures which are not based on dot products. In this first paper, we revisit neural networks built up of layers based on distance measures and Gaussian activation functions. These kinds of networks were only sparsely used in the past since they are hard to train when using plain stochastic gradient descent methods. We show that by using Root Mean Square Propagation (RMSProp) it is possible to efficiently learn multi-layer neural networks. Furthermore we show that when appropriately initialized these kinds of neural networks suffer much less from the vanishing and exploding gradient problem than traditional neural networks even for deep networks.
△ Less
Submitted 6 August, 2017;
originally announced August 2017.
-
The Effect of Wedge Tip Angles on Stress Intensity Factors in the Contact Problem between Tilted Wedge and a Half Plane with an Edge Crack Using Digital Image Correlation
Authors:
Seyedmeysam Khaleghian,
Anahita Emami,
Mohammad Yadegari,
Nasser Soltani
Abstract:
The first and second mode stress intensity factors (SIFs) of a contact problem between a half-plane with an edge crack and an asymmetric tilted wedge were obtained using experimental method of Digital Image Correlation (DIC). In this technique, displacement and strain fields can be measured using two digital images of the same sample at different stages of loading. However, several images were tak…
▽ More
The first and second mode stress intensity factors (SIFs) of a contact problem between a half-plane with an edge crack and an asymmetric tilted wedge were obtained using experimental method of Digital Image Correlation (DIC). In this technique, displacement and strain fields can be measured using two digital images of the same sample at different stages of loading. However, several images were taken consequently in each stage of this experiment to avoid the noise effect. A pair of images of each stage was compared to each other. Then, the correlation coefficients between them were studied using a computer code. The pairs with the correlation coefficient higher than 0.8 were selected as the acceptable match for displacement measurements near the crack tip. Subsequently, the SIFs of specimens were calculated using displacement fields obtained from DIC method. The effect of wedge tips angle on their SIFs was also studied. Moreover, the results of DIC method were compared with the results of photoelasticity method and a close agreement between them was observed.
△ Less
Submitted 6 January, 2015;
originally announced January 2015.
-
Image Processing Code for Sharpening Photoelastic Fringe Patterns and Its Usage in Determination of Stress Intensity Factors in a Sample Contact Problem
Authors:
Seyedmeysam Khaleghian,
Anahita Emami,
Nasser Soltani
Abstract:
This study presented a type of image processing code which is used for sharpening photoelastic fringe patterns of transparent materials in photoelastic experiences to determine the stress distribution. C-Sharp software was utilized for coding the algorithm of this image processing method. For evaluation of this code, the results of a photoelastic experience of a sample contact problem between a ha…
▽ More
This study presented a type of image processing code which is used for sharpening photoelastic fringe patterns of transparent materials in photoelastic experiences to determine the stress distribution. C-Sharp software was utilized for coding the algorithm of this image processing method. For evaluation of this code, the results of a photoelastic experience of a sample contact problem between a half-plane with an oblique edge crack and a tilted wedge using this image processing method was compared with the FEM results of the same problem in order to obtain the stress intensity factors (SIF) of the specimen. A good agreement between experimental results extracted from this method of image processing and computational results was observed.
△ Less
Submitted 6 January, 2015;
originally announced January 2015.
-
Design, Analysis, and Simulation of a Pipe-Welding Robot with Fixed Plinth
Authors:
Anahita Emami,
Seyedmeysam Khaleghian,
Mohammad Mahjoob Jahromi
Abstract:
Industrial requirements concerning the increased efficiency and high rate of manufacturing result in the development of manufacturer robots, and a vast group of these types of robots is used for welding. This study presented the design, analysis, and simulation of a pipe-welding robot with fixed plinth for a constant circular welding around the pipes. Design of a welding robot capable of kee** t…
▽ More
Industrial requirements concerning the increased efficiency and high rate of manufacturing result in the development of manufacturer robots, and a vast group of these types of robots is used for welding. This study presented the design, analysis, and simulation of a pipe-welding robot with fixed plinth for a constant circular welding around the pipes. Design of a welding robot capable of kee** the electrode orientation, welding speed, and distance between electrode and pipe surface constant can improve the quality of welding; thus, a five-linked articulated robot was designed for this purpose. Solving of direct and diverse kinematics and dynamics equations of the robot was done by means of Matlab software. The robot was also simulated using a program written in Matlab and the diagrams of angles, velocities, and accelerations of all the arms, and the applied force and torque of each arm required for drive the mechanism were obtained.
△ Less
Submitted 6 January, 2015;
originally announced January 2015.