Efe Tarhan

I am a Master’s student in Communication Systems at EPFL, supported by the Excellence Scholarship. My research interests are computer vision, foundation models, and 3D scene understanding.

I am currently working in the Visual Intelligence and Learning Lab (VILAB) under the supervision of Prof. Amir Zamir. Earlier, I worked in the Biomedical Imaging Group with Prof. Michaël Unser on multiresolution optimization and in the Image and Visual Representation Lab (IVRL) with Prof. Sabine Süstrunk on 3D scene editing. Before EPFL, I completed my B.Sc. in Electrical and Electronics Engineering at Bilkent University. During that time, I conducted research on visible light positioning under the supervision of Prof. Sinan Gezici for 2 years.

profile photo

Research & Selected Work

My earlier research was in computer vision and in detection and estimation. I am now interested in foundation models, multimodal data, 3D scene understanding, and diffusion models.

OmniDiMM: Bridging 1D, 2D, and 3D with Any-to-Any Multimodal Modeling
J. Toskov*, O. Barbany*, R. Singh*, J. Sakurai, Efe Tarhan, O. F. Kar, R. Bachmann, A. Zadeh, J. Allardice, C. Li, C. Torras, A. Dehghan, Amir Zamir
Neural Information Processing Systems (NeurIPS), 2026  
project page

A unified any-to-any multimodal model that bridges 1D, 2D, and 3D representations, enabling generation, retrieval, and object decomposition across text, images, and 3D modalities within a single framework.

Multimodality as Supervision: Self-Supervised Specialization to the Test Environment via Multimodality
Kunal Pratap Singh*, Ali Garjani*, R. Singh, M. U. Khattak, Efe Tarhan, J. Toskov, A. Atanov, O. F. Kar, Amir Zamir
International Conference on Learning Representations (ICLR), 2026  
project page

Test-Space Training (TST) pretrains vision models only on rich multimodal data from the environment they will be deployed in, instead of internet datasets, and asks whether scaling the modalities within a fixed test space can replace data generalism.

MiSensorKit: Multi-Modal Field Recorder for iOS Devices
Efe Tarhan, Kunal Pratap Singh, Amir Zamir
iOS Application, 2026
project page / App Store

An iOS app that records the camera, depth (LiDAR, or estimated on the phone), 6-DoF camera pose, IMU, GPS, 3D body skeleton, gaze, ambient light and air pressure in sync, frame by frame, with Python tools that turn recordings into training data and 3D reconstructions.

From Pixels to Wireframes: 3D Reconstruction via CLIP-Based Sketch Abstraction
Efe Tarhan, Filip Mikovíny, Eylül İpçi, Alberts Reisons
Course Project, 2025  
project page / presentation

Sketch abstraction, extended from 2D to 3D: Bézier curves are optimized through constrained Gaussian splatting, with Gaussians restricted to matte spherical blobs along each curve, so a few strokes form a wireframe-like sketch that captures an object's geometry.

IRS Aided Visible Light Positioning with a Single LED Transmitter
Efe Tarhan*, Furkan Kökdoğan, Sinan Gezici
Digital Signal Processing, 156, 2025  
repository / paper

An intelligent reflecting surface (IRS), optimized with the Cramér–Rao lower bound, lets a receiver be localized indoors with a single LED transmitter.


Website template by Jon Barron.