Batuhan Ozcomlekci
Hi, I'm Batuhan — a Machine Learning researcher based in Tübingen, Germany. I recently received my MSc in Machine Learning from the University of Tübingen.
I am working on the problems in the intersection of 3D Computer Vision, Graphics, and Machine Learning. For my MSc thesis in the RVH group, supervised by Prof. Gerard Pons-Moll, I worked on Generalizable 3D Human Gaussians.
I aspire to teach machines how to see and process the world beyond words and 2D images which are illusory proxies to the physical reality.
Education
M.S. Machine Learning
B.S. Computer EngineeringB.S. Physics (Second Major)Experience
Student Assistant
Computer Vision InternResearch Intern
Data Science Intern
Software Engineering InternSkills
Awards
DAAD StipendTÜBİTAK 2205 ScholarshipŞermin Savaşçı ScholarshipNews
Publications
Dimitrije Antić, Garvita Tiwari, Batuhan Ozcomlekci, Riccardo Marin, Gerard Pons-Moll
Proc. of the International Conf. on 3D Vision (3DV), 2024
Project Page / Paper / Code / Expand bibtex
@InProceedings{antic2024close,
author = {Dimitrije Antić and Garvita Tiwari and Batuhan Ozcomlekci and Riccardo Marin and Gerard Pons-Moll},
title = {CloSe: A 3D Clothing Segmentation Dataset and Model},
booktitle = {Proc. of the International Conf. on 3D Vision (3DV)},
year = {2024},
}Miscellaneous
SOMA-JAX
SOMA-JAX is a pure-JAX port of NVIDIA SOMA-X parametric human body model and procedural rig. The entire forward pipeline (identity blend, skeleton fit, FK, and LBS) and multi-stage pose inversion solver are packaged into a single fully differentiable, JIT-compiled graph supporting jax.jit, jax.vmap, and jax.grad across GPU, CPU, and TPU with sub-millimeter upstream parity.
Path Tracing Studio
An interactive WebGL GPU path tracing studio running directly in the browser, built with Three.js and custom GLSL shaders. Features progressive Monte Carlo rendering, adaptive GPU-timed sample batching for responsive interaction, and seamless raster previewing with real-time accumulation.
SMPL-JAX
SMPL-JAX is a pure-JAX library of fully differentiable, JIT-compiled SMPL and SMPL-X parametric human body models. Every operation works with jax.jit, jax.vmap, and jax.grad, enabling batched fitting and differentiable optimization on GPU/CPU/TPU, and it benchmarks faster than the PyTorch smplx reference.
Blender Data Generation Tool
A Blender tool in Python to output rendered images and masks with convenient camera parameters for Novel View Synthesis tasks employing 3D Gaussian Splatting and NeRFs. The tool’s mask rendering feature has been integrated into the Blendify framework as well.
Website template inspired by Michael Niemeyer and Maksim Zhdanov.