Hello, my name is Reza and I hold a bachelor's degree in electrical engineering
and have experience in computer vision and image processing for two years.
3D computer vision, autonomous vehicles, and medical image analysis are among my areas
of interest. It is however my goal to learn more about other areas of machine learning,
such as natural language processing and speech recognition.
Among my skills are some machine learning and data science tools such as
Python, Pytorch, Scikit-Learn, Pandas, Numpy, as well as visualization tools
such as streamlit, matplotlib, and seaborn. In terms of deploying tools, I am
familiar with Linux, Docker, FastAPI, and some other tools.
My passion for research has motivated me to pursue a master's degree in a related field, or
to join a research group as a volunteer.
Sep 2022 - Now remote, Sydney
Top Achievements/Task:
◾ Utilized Generative Adversarial Networks (GANs) for medical image-to-image translation, achieving an accuracy rate of 86%; Work was done by Pix2Pix modified model and this accuracy improved around 4% than previous model in the company.
◾ Annotate more than 2000 images for dental teeth detection (YOLO) and decay classification in coordination with specialists.
◾ Developed a CT-scan image segmentation solution and implemented a user-friendly web-face interface using FastAPI.
◾ Signal Processing for EEG signal analysis and achieving 96% accuracy in the DEAP dataset. Using pre-processing methods and combining ideas belong to model development.
◾ Carried out an ensemble learning approach to achieve 87% accuracy in segmenting Macular images.
◾ Maintain an annotation tool for specific object detection with Qt framework; This toolkit can be used as a faster object detection annotation.
Jan 2022 - Jun 2022 Iran, Tehran
Top Achievements/Task:
◾ Enhanced the Yolov7 algorithm by integrating an attention module (CBAM), resulting in a significant 5% accuracy boost in detecting smoke within natural environments (more information).
◾ Achieved a remarkable 91% accuracy in solving the face verification challenge and use optimization solutions to implement it edge devices like raspberry-pi.
◾ A multi-camera ball detection in volleyball videos with accuracy of 96% and detect players in each video; In this work a new detection algorithm designed to get a faster model as 50 FPS.
◾ Achieved 96% accuracy in ball detection for volleyball videos; in this work used an innovative detection algorithm for a faster processing speed of 50 FPS. (more information : link1, link2).
◾ Progressed and adopted a highly efficient depth estimation framework for car lighting nuances and shape analysis, achieving an impressive speed of 20 FPS.
Dec 2020 - Apr 2021 Iran, Tehran
◾ Digital Electronics Teaching Assistant (Winter2020)
Sep 2017 - Sep 2021 Iran, Tehran
Grade: 17.03/20
◾ Final Project: Implement CNN Algorithm on FPGA based VHDL
◾ Implement JPEG Compression method
◾ Digital Electronic TA
◾ Voice controlled Relay (with Nodemcu and Google Assistance)
◾ Implemet Digital Communication system in MATLAB
We used the TrackNet model and modified it in the pre-processing part and model before training it using the data we collected from YouTube videos.
In the trajectory portion of the project, I used 3D computer vision methods and algorithms to convert 2D points from different sides of the volleyball court into 3D points.
Ball detectio model (TrackNet)
3D trajectory mapping
In this project, deeplab model had been trained and the accuracy of segmentation was 87%
This project was an implementation of (A New Approach To Estimate Depth Of Cars Using A Monocular Image) paper but made a little modification in optimization part of this project. The idea was finding distance to each car from location of car (width, heght) and location of front and rear lights.
Issued by Iran's National Elite Foundation
Mar 2022
Our team achieved first place in the Rahneshan competition. In this project, we design a DOA estimation technique based on deep-learning models.
Our project has three-part: Design Antenna, Make Simulink model and Implement DeepLearning-based DOA estimation Of course my task was design a deep learning model and train a dataset.
we have two challenges in this project: Multi-Path and UWB signals, for solving these challenges I using Conv. neural network based on "Direction of arrival estimation in multipath environments
using deep learning" and "Direction of Arrival Estimation of Multiple UWB Signals" papers.