Yipeng Sun 孙熠芃

PhD Candidate in Computer Science

Pattern Recognition Lab · Friedrich-Alexander-Universität Erlangen-Nürnberg

My research lies at the intersection of inverse problems and deep learning, with a focus on medical imaging reconstruction and artifact compensation. I develop differentiable CT operators and GPU-accelerated pipelines for physics-informed training and test-time optimization. I also explore medical foundation models, LLM post-training, agentic systems, and generative models for image restoration.

Yipeng Sun

News

Feb 2026 Paper at BVM 2026 received top review scores — invited to submit to IJCARS Special Issue.
2025 Learning Wavelet-Sparse FDK accepted at Fully3D 2025.
2025 CBCT Motion Compensation paper accepted at iCT 2025.
2025 EAGLE loss paper published in SPIE Journal of Medical Imaging.
2025 Filter2Noise preprint released — self-supervised denoising for low-dose CT.

Research

I develop methods that are not only accurate but also interpretable and physically grounded, bridging the gap between deep learning and domain knowledge in medical imaging.

CT Image Reconstruction

Differentiable CT operators and end-to-end pipelines for physics-informed training. Trainable filters for filtered back-projection and GPU-accelerated reconstruction.

Image Denoising

Self-supervised and single-image denoising for low-dose CT under extreme data constraints, combining attention mechanisms with classical signal processing.

Motion Compensation

Generative model-based motion artifact compensation in cone-beam CT (CBCT). Test-time optimization with physics-constrained priors.

Differentiable CT Operators

Building CUDA-accelerated, differentiable forward/back projection operators that enable end-to-end optimization of reconstruction pipelines.

Selected Publications

Full list on Google Scholar.

Journal Articles

2025

EAGLE: An Edge-Aware Gradient Localization Enhanced Loss for CT Image Reconstruction

Y. Sun, Y. Huang, Z. Yang, L.-S. Schneider, M. Thies, M. Gu, S. Mei, S. Bayer, F.G. Zöllner, A. Maier

Journal of Medical Imaging, 12(1), 014001, 2025

Paper Code
2025

DRACO: Data-Driven Reconstruction Algorithm for Cone-Beam CT

C. Ye, L.-S. Schneider, Y. Sun, S. Mei, A.M. Kist, A. Maier

Physics in Medicine & Biology, 2025

Paper

Conference Proceedings

2025

Learning Wavelet-Sparse FDK for 3D Cone-Beam CT Reconstruction

Y. Sun, L.-S. Schneider, M. Gu, S. Mei, S. Bayer, A. Maier

Fully Three-Dimensional Image Reconstruction (Fully3D), 2025

Paper
2025

Compensating CBCT Motion Artifacts with Any 2D Generative Model

Y. Sun, L.-S. Schneider, M. Gu, S. Mei, S. Bayer, A. Maier

14th Conference on Industrial Computed Tomography (iCT), 2025

Paper
2024

Data-Driven Filter Design in FBP: Transforming CT Reconstruction with Trainable Fourier Series

Y. Sun, L.-S. Schneider, F. Fan, M. Thies, M. Gu, S. Mei, Y. Zhou, S. Bayer, A. Maier

International Conference on Image Formation in X-Ray CT (CT Meeting), 2024

Paper
2023

Compact Convolutional Transformers on Edge TPUs

Y. Sun, A. Kist

Bildverarbeitung für die Medizin (BVM) Workshop, pp. 141–146, 2023

Paper

Preprints

2025

Filter2Noise: Interpretable Self-Supervised Single-Image Denoising for Low-Dose CT with Attention-Guided Bilateral Filtering

Y. Sun, L.-S. Schneider, C. Ye, M. Gu, S. Mei, F. Wagner, S. Bayer, A. Maier

arXiv preprint, arXiv:2504.13519, 2025

Paper Code
2022

Deep Learning on Edge TPUs: A Review

Y. Sun, A. Kist

arXiv preprint, arXiv:2108.13732, 2022

Paper

Open Source

41

Eagle Loss

Edge-aware gradient localization enhanced loss for CT image reconstruction. Published in Journal of Medical Imaging.

PythonPyTorchCT
26

AutoCitation

Multi-agent tool that finds, verifies, and inserts real academic citations and BibTeX entries for LaTeX workflows.

PythonLLMMulti-Agent
18

Filter2Noise

Self-supervised single-image denoising for low-dose CT with interpretable attention-guided bilateral filtering.

PythonPyTorchDenoising
16

diffct

CUDA-accelerated differentiable CT operator library using Numba. GPU-parallel forward/back projection for deep learning reconstruction.

PythonCUDANumba

ConvNeXt Perceptual Loss

A PyTorch perceptual loss implementation based on the modern ConvNeXt architecture.

PythonPyTorchLoss Function

Diagnostic Devil's Advocate

Multi-agent AI system that challenges clinical diagnoses to prevent cognitive bias errors.

PythonMulti-AgentMedical AI

Experience

Jul 2023 – Present

Ph.D. Candidate & Project Leader

Pattern Recognition Lab, FAU Erlangen-Nürnberg

Advisor: Prof. Andreas Maier. CT/CBCT reconstruction, denoising, and motion artifact compensation. Leading AI software architecture for the BMBF-funded KI4D4E project (4D tomography), coordinating across 14 international partners.

Jul 2023 – Present

Researcher

Fraunhofer EZRT, Fürth

Deep learning for artifact compensation in industrial and scientific tomography. Robust training and inference pipelines for high-throughput CT systems.

Jun 2022 – May 2023

Research Assistant & Master's Thesis

Anki Lab, FAU Erlangen-Nürnberg

Neural architecture search for edge deployment. Genetic algorithm-based NAS and hardware-aware model design for Google Edge TPUs.

Education

Jul 2023 – Present

Dr.-Ing. in Computer Science

Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU)

Focus: AI in Medical Imaging · Supervisor: Prof. Andreas Maier

Sep 2020 – Jun 2023

M.Sc. in Medical Engineering

Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU)

Sep 2015 – Jun 2019

B.Eng. in Measurement and Control Technology

Nanjing University of Science and Technology (NJUST)

Honor Graduate

Awards

2026 BVM 2026 — Paper received top review scores, invited to submit to IJCARS Special Issue
2019 Honor Graduate, Nanjing University of Science and Technology
2015–19 University Scholarships (Multiple Awards), NJUST

Contact

Email

yipeng.sun@fau.de

Address

Pattern Recognition Lab
Friedrich-Alexander-Universität
Erlangen-Nürnberg
Martensstraße 3, 91058 Erlangen, Germany

Profiles

Google Scholar
GitHub
LinkedIn
FAU Profile