Efe Tarhan

Efe Tarhan

MSc Student in Communication Systems @ EPFL

Projects

Large-Scale 3D Scene Relighting using Pre‑Trained Diffusion Models

3 minute read

Date:

This project is a NeRF editing framework that enables localized relighting and texture edits using pretrained diffusion models. Built on the DDS pipeline, it integrates wavelet-based gradient filtering to preserve reflections and fine appearance details during editing. The framework keeps scene geometry fixed after NeRF training, ensuring structural consistency, and enhances edits with surface normal prediction for improved view-dependent rendering. The result is high-fidelity, semantically guided 3D scene edits with strong reflection and color consistency. The GitHub repository can be found in the following link.

Cover image for Large-Scale 3D Scene Relighting using Pre‑Trained Diffusion Models
Cover image for DetoxText: Text Detoxification Using Finetuned Encoder-Decoder Models
Cover image for From Pixels to Wireframes: 3D Reconstruction via CLIP-Based Sketch Abstraction

Wireless Train Signalization

3 minute read

Date:

This project is developed as part of the Bilkent University Electrical and Electronics Engineering Industrial Design Project. It aims to introduce a wireless railway signalization system utilizing RF-based wireless technologies and passive RFID antenna balises. This system is designed to optimize track capacity and safety by providing continuous communication and real-time management of train movements.

Cover image for Wireless Train Signalization

Machine Learning-Enhanced Visible Light Positioning in IRS-Assisted Indoor Environments with a Single LED Transmitter

1 minute read

Date:

This study explores visible light positioning (VLP) using a single LED transmitter and an intelligent mirror array in confined spaces with obstructed paths. Evaluated algorithms include Maximum Likelihood Estimation (MLE), K-Nearest Neighbors (KNN) Regression, and Neural Networks. MLE showed superior accuracy, while intelligent reflecting surface orientations enhanced receiver localization.

Cover image for Machine Learning-Enhanced Visible Light Positioning in IRS-Assisted Indoor Environments with a Single LED Transmitter

IdentiFusion: A Multimodal Approach to Facial Attribute Recognition using FaceBERT and LightenedMOON

2 minute read

Date:

IdentiFusion introduces a multimodal system combining NLP and Computer Vision to match textual face descriptions with facial images. It integrates FaceBERT for extracting facial features from text and LightenedMOON for recognizing facial attributes from images. This project showcases effective NLP and Computer Vision integration for image retrieval, matching, and recommendation systems, aiding law enforcement and forensic investigations. 

Cover image for IdentiFusion: A Multimodal Approach to Facial Attribute Recognition using FaceBERT and LightenedMOON

Monkeypox Disease Detection using Classical Backbone Models

less than 1 minute read

Date:

This project, under ICONLAB, aims to detect and distinguish monkeypox from smallpox, chickenpox, and measles. It focuses on raising awareness in the global medical imaging community. For accurate classification, we utilized deep learning backbone models such as VGG16, Inception Net, and ResNet.

Cover image for Monkeypox Disease Detection using Classical Backbone Models

Flight Price Prediction using ML Techniques Written From Scratch

3 minute read

Date:

This project explores the relationship between airline ticket prices and various factors such as airline, destination, flight times, booking period, and flight class. We developed custom-built machine learning models using fundamental Python libraries (numpy, pandas) to predict ticket prices based on these features.

Cover image for Flight Price Prediction using ML Techniques Written From Scratch

Disease Detection from X-Ray Images with Classical Backbone Networks

less than 1 minute read

Date:

This TUBITAK project aims to develop a system for detecting diseases from X-ray images using neural networks. It involves preparing and labeling a dataset, preprocessing images to 224x224 pixels, and binarizing labels for classification. Data augmentation techniques like rotation, zoom, and shifts enhance the training dataset.

Cover image for Disease Detection from X-Ray Images with Classical Backbone Networks

Whack-A-Mole-Game with VHDL on Basys3 Board

1 minute read

Date:

The primary goal of this project is to develop a digital game, Whack-A-Mole, using a VGA connector and a BASYS3 FPGA. The project aims to enhance digital design abilities, focusing on debugging, glitch removal, and the implementation of sequential and combinatorial design elements to create a glitch-free game.

Cover image for Whack-A-Mole-Game with VHDL on Basys3 Board