Салех Хади Мухаммед

Monocular Depth Estimation Based on Active Learning

2026 · CHAPTER · en

Estimating depth is a necessary task to understand and navigate the environment surrounding us. Over the years, many active sensors have been developed to measure depth, but they are expensive and require additional space for mounting. A cheaper alternative is to estimate depth from a single RGB image taken by an ordinary monocular camera, which can be placed even inside the smartphone. However, it is a well-known problem that neural networks require huge amount of labeled data to be effectively learned. That fact serves a barrier to the further development of the monocular depth estimation. In this paper, we address this problem. We propose a novel active deep learning training framework that reduces the dataset volume ratio by adaptively selecting the most informative data for labeling that focus on the most relevant human vision features for monocular depth estimation, which help us identify the image pixels that are most relevant for depth estimation. Our methodology indicates that it is possible to reduce the amount of labeled training data by 81% and at the same time preserve the comparable accuracy on the KITTI Odometry dataset.

Выполнение распределенных вычислительных экспериментов на MLOps платформе НИУ ВШЭ

2025 · ARTICLE · ru

Несмотря на распространение и успешные применения средств интеллектуального анализа и обработки данных для решения отдельных прикладных задач, все еще не решена проблема разработки технологии создания таких программных средств. В работе в контексте единого процесса MLOps создания технологий машинного обучения рассматриваются возникающие задачи автоматизации и выполнения распределенных вычислительных экспериментов на базе единой вычислительной платформы. Разрабатываемая в НИУ ВШЭ платформа MLOps предназначена для развертывания интеллектуальных веб-сервисов и программных средств анализа данных. Платформа должна управлять доступными локально и в облачной среде разнородными ресурсами и объединять их с ресурсами вычислительного кластера cHARISMa НИУ ВШЭ под управлением Slurm. Таким образом актуальна задача интеграции указанных ресурсов для проведения вычислительных экспериментов, реализации конвейеров настройки моделей машинного обучения, решения задач обработки и анализа данных. Особенностями решаемой задачи являются рассмотрение процесса вычислений, как составной части технологии создания интеллектуальных веб-сервисов, обусловленная этой технологией необходимость использования разнородных ресурсов и использование единой гибридной платформы для выполнения вычислений. В работе предложено решение указанной задачи интеграции вычислений и приведены результаты апробации решения для интеллектуальных веб-сервисов. Показана принципиальная возможность такой интеграции разнородных ресурсов в одном вычислительном эксперименте на базе расширяемой пользователем объектной модели эксперимента и предметно-ориентированного языка его спецификации, решены вопросы динамического управления развертыванием интеллектуальных приложений, интеграции конвейеров обработки данных, веб-сервисов и наборов данных для выполнения распределенных вычислительных экспериментов.

2023 International Symposium ELMAR, 11-13 September 2023, Zadar, Croatia

2023 · BOOK · en

Estimating depth is necessary to understand and navigate the environment surrounding us. Over the years, many active sensors have been developed to measure depth, but they are expensive and require additional space for mounting. A cheaper alternative is estimating depth from a single RGB image taken by an ordinary monocular camera, which can be placed inside the smartphone. However, state-of-the-art depth estimation algorithms are based on complex deep neural networks that are too slow for real-time inference on mobile platforms which can be mounted, for instance, on a micro aerial vehicle. That fact is a barrier to the further development of monocular depth estimation. In this paper, we address this problem. We utilize recent advancements in the architecture of lightweight networks to reduce complexity. We propose a novel lightweight network design with competitive accuracy and significant complexity reduction compared to existing approaches. Our methodology indicates that it is possible to achieve inference speeds accelerated by an order of magnitude on NVIDIA Jetson Nano and, at the same time, preserve the comparable accuracy on the KITTI Odometry dataset in comparison with the current state-of-the-art algorithms.

An Ontology Model to Facilitate Sharing Risk Information and Analysis in Construction Projects

2023 · CHAPTER · en

Construction projects face a high level of dynamic and various risks. Risks may result in deviation from pre-determined construction project’ objectives. Systematic risk analysis is critical for sharing risk information related to decision making thus effective risk management. In this study, a Risk Analysis Ontology RA-Onto is proposed that may facilitates development of databases and information sharing for risk analysis. A detailed review of the literature on construction risks has been carried out to development of the RA-Onto that organizes risk knowledge into unified classes together with corresponding properties and relations. Ontology is evaluated theoretically and practically by using five case studies. RA-Onto could be used to support decision-making during the risk management. It enables companies to corporate memories, create databases, and develop a model to support the systematic risk analysis for better decision making.

Arabic Text-to-Speech Service with Syrian Dialec

2023 · CHAPTER · en

This research aims to develop an Arabic text-to-speech (TTS) service with Syrian dialect, which is a variety of Arabic spoken in Syria and some neighboring countries, with easy access to it for people with disabilities or difficulty reading Arabic, such as people with visual impairments or learning disabilities. To achieve this goal, we employ two state-of-the-art Machine Learning (ML) approaches: Tactron2 and Transformers, which have achieved impressive results in various natural language processing tasks, including TTS. We compared the two approaches and evaluated the resulting TTS service using subjective measures. Our results show that both approaches can produce high-quality speech in the Syrian dialect, but transformers have the advantage of being more efficient and more flexible in handling different languages and accents.

Efficient Monocular Depth Estimation for Edge Computing Platforms

2023 · CHAPTER · en

Estimating depth is necessary to understand and navigate the environment surrounding us. Over the years, many active sensors have been developed to measure depth, but they are expensive and require additional space for mounting. A cheaper alternative is estimating depth from a single RGB image taken by an ordinary monocular camera, which can be placed inside the smartphone. However, state-of-the-art depth estimation algorithms are based on complex deep neural networks that are too slow for real-time inference on mobile platforms which can be mounted, for instance, on a micro aerial vehicle. That fact is a barrier to the further development of monocular depth estimation. In this paper, we address this problem. We utilize recent advancements in the architecture of lightweight networks to reduce complexity. We propose a novel lightweight network design with competitive accuracy and significant complexity reduction compared to existing approaches. Our methodology indicates that it is possible to achieve inference speeds accelerated by an order of magnitude on NVIDIA Jetson Nano and, at the same time, preserve the comparable accuracy on the KITTI Odometry dataset in comparison with the current state-of-the-art algorithms.

Robust Collision Warning System based on Multi Objects Distance Estimation

2022 · CHAPTER · en

The annual number of road deaths is still increasing, especially in less developed and developing countries. Road accidents are the 5th cause of death and the leading reason for death among young people between 5 and 29 years of age in 2030. In this study, a robust solution is implemented by integrating object recognition with distance estimation to maximize driving safety. The proposed system will be able to detect common objects within the region of interest on the road and estimate how far these objects are from the camera position. The system will trigger an alarm to attract the driver’s attention in real time when the distance to one of the detected objects is less than a predefined threshold value. In this work YOLO (You Only Look Once) approach is used to detect the objects in real time and the properties of the depth map based on deep learning is applied to estimate the distance at a given point.

Machine Learning-Based Crime Prediction

2022 · CHAPTER · en

Eolang: Toward a New Java-Based Object-Oriented Programming Language

2021 · CHAPTER · en

Object-oriented programming (OOP) is one of the most common programming paradigms used for building software systems. However, despite its industrial and academic value, OOP is criticized for its high complexity, low maintainability and lack of rigorous principles. Eolang (a.k.a. EO) was created to solve the above problems by restricting its features and suggesting a formal object calculus for this programming language. This paper seeks to analyze the Eolang language and compare it to other OOP languages in order to develop the core features of this new language.

Exploring the Eolang-Java Integration and Interoperability

2021 · CHAPTER · en

In recent times, the subject of interoperability has become very popular. In large-scale software applications development, it is a common practice to combine multiple languages in solving peculiar problems and developing robust solutions. The ability to combine multiple languages allows an easy migration of an existing project from one language to another or use existing libraries in another language. This makes interoperability a force to be reckoned with when developing new programming languages. The Eolang programming language is a new research and development initiative aimed at achieving true Object-Oriented Programming by having all components of the program as objects. As such, the construct and syntax of Eolang is vastly different from that of Java. This makes integration and interoperability between these two languages a challenging issue related to method/object naming conventions, keywords and operators, etc. In this paper we explore the potential of Eolang interoperability with Java by looking at the interoperability mechanisms of some other languages with Java, describe ways to overcome these challenges with Eolang and develop the solution. Specifically, we focus on the possibility to call Java code from Eolang while the semantics of both languages remain preserved. Our solution allows Java code to be called in Eolang through wrappers that turn Java classes and methods into Eolang Objects.