Sharif Digital Repository

One of the most important applications of artificial intelligence, and especially deep learning is image captioning. Given an image, the task is to automatically produce a sentence, describing the image. Image captioning has several real world applications like helping the blind understanding the images, generating automatic captions for the social media, etc. In the past, several different methods for image captioning have been used, but after the emergence of deep learning, like many other areas, image captioning algorithms have been improved significantly. In this thesis, I talk about a specific method for image captioning, called ”Deep Compositional Captioning. In this method, at first... 

Detection text in natural images is a challenging task due to the complex backgrounds in an image. complex backgrounds, changes in ambient light, changing viewing angles, and other factors can make systems difficult to detection text. Hence text detection is always an problem. Since detection and recognizing a text in an image has many uses such as translating texts for tourists, helping the blind, etc., recognizing a text in different languages is important. In this thesis, we first examine the three methods of Reading Text in the Wild with Convolutional Neural Networks and FOTS and CRAFT. Then we prepared two Persian data sets. The first data set contains images to which Persian texts have... 

Image retrieval refers to the task of finding images related to a query image within an image set. Due to ever-increasing volumes of data, it has become increasingly necessary to find suitable and efficient methods for searching in massive databases. In this thesis, modern image retrieval techniques developed within the last 15 years have been studied, with an aim to satisfy three primary constraints of efficiency, accuracy, and low memory usage. Our focus has been on content-based retrieval; meaning that instead of using text and other information, we directly utilize image features for analysis and processing. To achieve this, we studied two established techniques, the bag-of-words model,... 

Face recognition, which is one of the most important biometrics, has always been one of the main challenges in many security issues, such as verifying the identity of customers of financial institutions and passengers at the airport, and such issues have many applications in daily life. Face recognition has always been an important issue in computer vision and pattern recognition. Currently, several methods based on deep networks have shown great results in face recognition, among which the following can be mentioned.1.The deep face was introduced by Facebook in 2014; 2.Face-net was presented by Google in 2015 ;3.VGGFace was presented by Oxford University in 2015; 4.Openface was presented by... 

Non-overlapping Cameras in multi-camera systems have become prevalent in robotics and computer vision research; therefore, it is possible to cover the wide field of view, and researches have been done for computing extrinsic parameters of cameras. These cameras do not have any overlap in their views, so obtaining the corresponding point in their images is somehow impossible. Light and shadow geometry is analogous to Structure from Motion problem. In this thesis,we study Structure from Motion problem and have tried to propose an approach for estimating extrinsic parameters of non-overlapping cameras in Multi-camera systems. We formulate the problem by using light and shadow geometry and... 

Quantum mechanics is very succesful in the description and prediction of the microphysical world. So nowadays very few doubt its authenticity. Nevertheless, there are some conceptual issues that are still under debate, such as action at a distance. It took a while after the introduction of the EPR paradox in 1935 until John Stewart Bell derived an inequality, using local hidden variables. This inequality was in contrast with predictions of quantum mechanics. Experiments also contradicted this inequality. This implies that there is action at distance which, in turn suggests, the violation of the special relativity. In this thesis, we consider this issue and shall see that on the one hand we... 

Nowadays, with the increasing size of data,it’s impossible to collect data and fast classification by human, and needs for an automated classification and data analysis, is more interested. Data classification is a process of giving the training data along with their class labels to the learning agent, which learns the relation between the instances and the labels. Then make a prediction to the label of the training data.In this thesis we will observe the classification of the multi-label data. Multi-label data have more than one label. In other words, each instance appears with a vector of labels.In this thesis, a method based on nearest neighbor is proposed to classify the multi-label... 

Credit scoring is a mechanism used to quantify the risk factors relevant for an obligor’s ability and willingness to pay. Credit scoring has become the norm in modern banking, due to the large number of applications received on a daily basis and the increased regulatory requirements for banks. The meaning of credit scoring is to assign scores to the characteristics of debt and borrowers, historical default, and other loss experienced as an indication of the risk level of the borrower. The aim of the credit score model is to build a single aggregate risk indicator for a set of risk factors.
In this study, current method in which Maskan bank will give a loan is explained. We developed a... 

Recent progress in developing Artificial Neural Network (ANN) meta-models has paved the way for reliable use of these models in the prediction of air pollutant concentrations in urban atmosphere. However, improvement of prediction performance, proper selection of input parameters and model architecture remain key challenges to their practical use. This study’s main objectives are: selecting proper input parameters for ANN meta-models, optimizing the ANN models to achieve the most accurate hourly prediction for a case study (City of Tehran), coupling the ANN method with filtered input data and evaluating its privilege. In the current study the ANNs were constructed to predict criteria... 

In this thesis, static and dynamic analysis of nano beams based on second strain gradient theory is presented. Due to their small sizes, nano electro mechanical devices (NEMS) hold tremendous promise for novel, versatile and very sensitive devices for different applications ranging from actuators, transducers and also mass, force, light and frequency detectors. Therefore accurate modeling and analysis of such devices has an important role in their design and performance improvement. Neglecting the size effect, traditional theory of elasticity can not be suitable to predict mechanical behavior of these systems and so, it should be used non-classical theories which include size dependency... 

It has been about a decade that cosmological datas have shown the universe is accelerating. To describe this acceleration, cosmologosts use cosmological constant as the easiest way to describe,or the modified gravity models or smooth dark energy models as other ways. In this thesis we solved a specific structure formation problem, density contrast, for linear structures, in two different methods. the first method was LCDM model and the second method was a specific f (R) gravity model. We shpwed that these two solutions are about the same.
 

In this thesis, we study the task of sign language translation using a non-end-to-end framework by utilizing large language models. The proposed method initially generates a sequence of glosses corresponding to a sign language video through a continuous sign language recognition model, which is then considered as input for a language model. The role of a language model in this framework is to translate the sequence of glosses into coherent sentences in spoken language. Given the limited training resources available for machine translation models on the specific problem of sign language to text translation, leveraging large language models can be highly beneficial due to the accumulated...