Atakan Topaloğlu

I'm a first-year MSc student in Electrical Engineering and Information Technology at ETH Zürich, specializing in 3D Scene Understanding and Visual Representation Learning. I work as a Research Assistant in the Photogrammetry and Remote Sensing Lab, supervised by Prof. Konrad Schindler.

I completed my BSc at Koç University and studied at Robert College in Istanbul. During my undergraduate studies, I spent three years at Siemens alongside highly talented colleagues as an R&D Working Student.

For any inquiries, feel free to reach out to me via mail!

CV Mail Scholar Github LinkedIn

Awards:

In 2024, I was awarded the IEEE Signal Processing Society Scholarship as the first ever recipient from a Turkish University. In 2025, the scholarship was renewed during my first year of MSc in ETH Zürich.
We were awarded the best graduation project award on our work on Single-Image-Super-Resolution Evaluation Methodologies under the supervision of Prof. Murat Tekalp.
I gratefully thank the Hisar Educational Foundation for supporting my undergraduate studies with their scholarship.

📚 Publications

OracleGS: Grounding Generative Priors for Sparse-View Gaussian Splatting
Atakan Topaloglu, Kunyi Li, Michael Niemeyer, Nassir Navab, A. Murat Tekalp, Federico Tombari
Proc. of the IEEE Winter Conf. on Applications of Computer Vision (WACV), 2026
Project Page / Paper / Supplementary /

@InProceedings{topaloglu2025oraclegsgroundinggenerativepriors, 
	author = {Atakan Topaloglu and Kunyi Li and Michael Niemeyer and Nassir Navab and A. Murat Tekalp and Federico Tombari}, 
	title = {OracleGS: Grounding Generative Priors for Sparse-View Gaussian Splatting}, 
	booktitle = {Proc. of the IEEE Winter Conf. on Applications of Computer Vision (WACV)}, 
	year = {2026}, 
}

Image-Difficulty-Aware Evaluation of Super-Resolution Models
Atakan Topaloglu*, Ahmet Bilican*, Cansu Korkmaz, A. Murat Tekalp
Proc. of the IEEE International Conf. on Image Processing Workshop (ICIPW), 2025
Paper / Video / Poster /

@InProceedings{topaloglu2025imagedifficultyawareevaluationsuperresolutionmodels, 
	author = {Atakan Topaloglu and Ahmet Bilican and Cansu Korkmaz and A. Murat Tekalp}, 
	title = {Image-Difficulty-Aware Evaluation of Super-Resolution Models}, 
	booktitle = {Proc. of the IEEE International Conf. on Image Processing Workshop (ICIPW)}, 
	year = {2025}, 
}

* denotes equal contribution

💻 Selected Projects from Siemens R&D

3D Semantic Segmentation with SAM2 Memory Bank

We noticed that GaGa (Lyu et al.) repeatedly labeled the same objects when revisiting regions, especially textureless regions, such as walls, failing to maintain consistency across views. To address this, we leveraged the existing SAM2 memory bank and used our dense images and pose-based image warping to project previous masks onto new viewpoints. This simple approach enforced mask consistency in indoor scenes with "loop closure", reducing duplicate labeling especially in textureless regions.

Request Access via Form

3D Gaussian Splatting from Smartphone Cameras

We found that aggressive downsampling and color palette quantization of input images had minimal effect on Gaussian Splatting quality in indoor scenes. Leveraging this insight, we redesigned the data pipeline to apply these steps after structure-from-motion (SfM), reducing memory consumption by 83%. In addition, a lightweight interface was developed to make the pipeline accessible for everyday captures from smartphone cameras, lowering the barrier for SMEs and non-expert users to generate their own splats. The final model was reconstructed with maximum GPU VRAM utilization of 3.9 GB from 729 images, occupying 122 MB on disk.

Request Access via Form

Building Information Modelling (BIM) PoC on CPU

Developed a proof-of-concept for a Building Information Modeling viewer designed to run efficiently on consumer-grade hardware. This demonstration runs smoothly on an 8th Gen Intel i5 CPU (4 cores, 8 GB RAM, 2.8 GHz) with no dedicated GPU.

Request Access via Form