I'm currently a last-year undergraduate student in Computer Engineering at Sharif University of Technology (the best university in Iran, according to QS ranking 2023). My research interests are among intersection of these areas: 1) 3D Computer Vision, 2) Computer Graphics, and 3) Robotics. I think that combination of these areas, will make a new revolution in our lives, bringing us Fully Autonomous Vehicles, Assistant Robots and replacing smart phones woth smart glasses. All of these will happen if we can make the world perceptible for machines, going further than human perception! I am continuously seeking researches related to these subjects and have done some projects untill now, which brought me a lot of learning opportunities.
AmirHossein Razlighi
arazlighi@gmail.com
amirhossein.razlighi@sharif.edu
Cumulative GPA (untill now): 19.21/20.0 | Major GPA (untill now): 19.5/20.0
GPA: 19.86/20.0
( This job is related to my university CO-OP program )
I worked on a new project named "Driver Arrived AR" that aims to locate and show drivers in
"Augmented Reality" mode in webapp.For this project, I worked with tensorflow models and
different CNN architectures to do "Realtime Object Detection" for detecting cars and normalizing
location of user (blue-dot).
I also researched on articles about "Realtime Semantic Segmentation" and therefore, read CNN
architectures such as"Refine-Net/U-Net/DeepLab/DeeplabV3plus" and tried to develop them using
tensorflow (training phase) and tensorflow.js (deployment on web).
I also made different communications with other departments (AI - Design - etc.) to instantiate
a team named "Tapsi-Lab" that its main goal is doing R&D on Augmented Reality and Computer
vision.
Working on the subject of Vision-Based Robotic Manipulation. In this experience, I worked with multimodal models and got familiar with vision-based methods in robotics.
Working on the subject of 3D Reconstruction using Deformable NeRF models. In this experience, I am working with a wide variety of SOTA computer vision methods and computer graphics methods.
(This job is not contract based and is voluntarily)
I Worked on weakly supervised semantic segmentation
Under supervision of Professor Shohreh. Kasaei
I am continuously learning and seeking new opportunities to gain experience and cooperate with experts. I had the honor of working with great people and learning from them until now and I gained many skills which I hope I can improve them more in the future.
I continuously seek opportunities to share my knowledge and learn more from interacting with people who are active in the field.
One of the best societies to be active and learn from, is undergrad students. I became TA in different roles and different courses
to make my knowledge from these courses more complete and learn more and more by designing homeworks and answering questions of students.
Some of the most important activities of mine in this area are:
I also have many voluntarily activities in the different societies and seminars in the Computer Engineering Department. Some of
the most important ones are:
This is a PyTorch implementation of tiny NeRF (Neural Radiance Field). The original paper is NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis. This version is a much smaller version of the original model, which you can simply train even on your cpu(!) in a reasonable time.
In this project, I implemented basic ray tracing concepts from scratch in python, using only numpy and matplotlib. The goal is to understand the basic concepts of ray tracing, and to have a working code that can be used to test and experiment with different ideas.
This is a simple implementation of marching squares algorithm in Computer Graphics, in python. The code main.py creates 2d terrains using the algorithm and shows them as a plot. The code contains different value systems for nodes of squares (binary as 0 and 1, and float as 0.0 to 1.0) and different ways of assigning them values (using simple uniform random generator, or using different kinds of noise). The code also contains a simple implementation of some noise generators.
This project is aimed to build a search engine for Semantic Scholar. This search engine is implemented in 3 phases. The first phase aims to build the index, compression of index and doing basic retrieval tasks using TF-IDF. The second phase aims to do the retrieval tasks using machine learning methods. We investigate Naive Byaes Methods alongside Neural Networks models and also Language Models to rank the documents(learning to rank). The third phase scraps the data from semantic scholar website and classifies that with respect to the author. Then we update the index using this new data and then, we build a frontend system using Streamlit, to make the user experience of the search, easier.
PortSet is a dataset for focusing on foreground objects in an image (like Portrait mode in phones!). This dataset is created so that you can train models which can detect background objects, foreground objects or even learn to directly blur the unimportant stuff in an image!
Introducing a new loss function for comparing the difference between two blurred images in an accurate way.
This project aims to create a cminus compiler to compile CMinus codes (with specific grammar) to python code. It has 4 phases which contain definition of scanner, parser, error analysis and semantic analysis.
A simple and fast webApp for day-to-day communication. This repo is for YumGram! which is a web-based messenger application with many customizable features for ease of use.