Mohammadreza Davoodi Behbahani

Mohammadreza Davoodi Behbahani
AI Developer and Researcher

I am currently a non-profit researcher in the Data Science and Machine Learning Laboratory.
I hold an MSc in Artificial Intelligence from Iran University of Science and Technology, specializing in Image Processing and Computer Vision. Prior to that, I earned a Bachelor’s degree in Electrical Engineering with a focus on Communication from Isfahan University of Technology.
Passionate about designing and implementing smart systems, I enjoy working on innovative solutions that integrate AI and advanced technologies.

Download CV

IELTS Result: 7.0 Overall: {Listening:7.5, Reading:7.5, Writing: 6.0, Speaking: 7.0}

Email

me[AT]mrdavoodi[DOT]ir

Phone

(+98) 937-029-0064

Age

28

LinkedIn

MrDavoodi

Education

Masters in Computer Engineering (Artificial Intelligence)

September 2019 – February 2023

Thesis Title: One-shot Similarity on Image Recognition
Iran University of Science and Technology (4th ranked in Iran according to QS)

Bachelor in Electrical Engineering

September 2015 – September 2019

Isfahan University of Technology (5th ranked in Iran according to QS)

Research Projects

Deepfake Detection in Videos

– Developed a novel deepfake detection pipeline by Combining EfficientNet and Vision Transformer,
GenConvViT, Altfreezing, and UCF methods.
– Created a custom dataset, applying a deepfake technique to generate realistic manipulated videos.
– Achieved 98% accuracy with UCF on my custom dataset.
– Explored Ensemble Learning as another approach to solve deepfake detection.
– Tested on Qwen2-VL as a potential future research.

Multimodal Analysis of Movie Data (Ongoing Project)

– Performing multimodal analysis on movie data from different sources
– Applying data preprocessing, feature extraction, and statistical analysis across different data modalities.
– Aiming to provide insights for predictive modeling of content ratings using multimodal signals.

Logo Detection Using YOLOv5

– Collected a custom dataset of 100 logo categories, including brand and TV channel logos extracted
from internet and TV program screenshots then labeled using labelImg.
– Fine-tuned YOLOv5 for multi-class logo detection.
– Achieved over 80% detection accuracy, demonstrating the effectiveness of deep learning for visual logo detection.
– Explored challenges of dataset imbalance and variability in real-world logo appearances.

Retrieval-Augmented Generation (RAG) Service with LLMs

– Designed and implemented a document-based question answering service using a lightweight large language model (Qwen-2B).
– Integrated a retrieval pipeline to fetch relevant document passages and augment the LLM’s responses for improved factual accuracy.

One-shot Similarity on Image Recognition (MSc. Thesis)

– Finetuned a pretrained YOLOv5 network (ImageNet classification weights) for image recognition on 16 classes.
– Implemented similarity method for 4 additional classes, using only a single image per class as the training sample.
– Explored fine-tuning different number of layers and their effect on accuracy.

Skills

Python

Pytorch, Tensorflow, Opencv, Numpy, Scipy, Scikit learn, Matplotlib, Pandas

C and C++
Matlab
Linux
Git
Docker

Research Interests

These are the areas I’m interested to work.

Deepfake Detection and Multimedia Forensics
Computer Vision and Image Processing
Natural Language Processing (NLP) and Text Understanding
Large Language Models (LLMs) and Generative AI
Vision-Language Models (VLMs) and Multimodal AI
Applied Machine Learning and Deep Learning
Trustworthy and Responsible AI