Curriculum Vitae
Professional Summary
- Generative AI Engineer and Applied Scientist with a PhD in Computer Science, over six years of experience in computer vision and human pose tracking, and two years of hands-on work in diffusion-based generative model development.
- Deep research expertise in multi-person pose tracking, covering subdomains such as person detection, object tracking, pose estimation, person search, and re-identification.
- Proven track record in building large-scale datasets (e.g., PoseTrack21) and training deep learning architectures, including convolutional and transformer-based models.
- Skilled in designing and deploying end-to-end generative AI pipelines, with a focus on high-fidelity digital human synthesis for production environments.
- Extensive experience with training strategies for resource-constrained environments, including LoRA, mixed precision, quantization, and gradient accumulation.
- Strong command of modern GPU training environments, MLOps workflows, and cross-functional collaboration in fast-paced teams.
Experience
-
7. 2023 - Present AI Engineer, Technical Lead
Verce GmbH, Cologne, Germany - Led the creation and strategic direction of the in-house AI team.
- Developed and optimized generative image pipelines for digital fashion models using state-of-the-art diffusion techniques.
- Built scalable training and evaluation systems for fine-tuning models including SD1.5, SDXL, SD3, and Flux.
- Introduced efficient training strategies (LoRA, Mixed Precision, Gradient Accumulation) for low-resource GPU environments.
- Deployed internal AI tools within dockerized Linux environments, supporting the CGI production workflow.
- Maintained and administered the company GPU server stack (StackIT Cloud, CUDA, driver updates, GPU scheduling).
-
4. 2018 - 6. 2023 Research Assistant
Computer Vision Group, University of Bonn, Germany - Conducted PhD research on multi-person pose tracking in videos.
- Created and annotated the large-scale PoseTrack21 dataset for person search, multi-object tracking, and multi-person pose tracking.
- Published at top conferences including CVPR, ECCV, BMVC, AAAI, and NeurIPS.
- Maintained lab GPU infrastructure and supervised multiple Bachelor and Master theses.
- Led developments in person detection, re-identification, temporal tracking, and spatial consistency.
-
10. 2015 - 12. 2017 Student Research Assistant
Computer Vision Group, University of Bonn, Germany - Worked on 2D and 3D human pose estimation tasks using deep learning.
- Assisted in research and tool development for pose datasets and model evaluation.
-
2. 2011 - 9. 2013 Working Student
OwnSoft GmbH, Cologne, Germany - Co-developed management software for a multimedia company in C# and .NET.
Academic Interests
-
Generative AI
- Training and fine-tuning diffusion models for high-resolution human image generation
- Improving realism, identity consistency, and fidelity in digital human synthesis
- Exploring novel architectures and training techniques for generative models
-
Human Pose Estimation and Tracking
- Multi-person pose estimation and tracking under occlusions and motion
- Person re-identification and long-term temporal association in videos
- Person detection, search, and tracking with spatial-temporal attention mechanisms
- Dataset design and annotation workflows for large-scale pose tracking benchmarks
-
Deep Learning and Computer Vision
- Designing and benchmarking CNN and Transformer-based architectures
- Efficient training strategies including quantization, mixed precision, and distributed training
- Interpretability and robustness in model predictions
-
Infrastructure and Tooling
- GPU cluster management, job scheduling, and performance debugging
- Docker-based deployment pipelines for reproducible training and inference
Skills
-
Programming Languages
- Python (advanced), C++, Bash
-
ML & Deep Learning Frameworks
- PyTorch, Diffusers, Transformers, Accelerate, PyTorch Lightning
-
Computer Vision & Data
- OpenCV, TorchVision, NumPy, Matplotlib, numba, TensorBoard
- Dataset creation, annotation tooling, image/video pre-processing
-
Training Techniques
- Mixed precision, quantization, gradient accumulation, LoRA
- Fine-tuning, transfer learning, model compression
-
Infrastructure & MLOps
- Docker, Make, Git, Linux/Ubuntu, GPU scheduling and driver stack
- ComfyUI, StackIt Cloud, Microsoft Teams, Jira, Miro
-
Tools & IDEs
- NeoVim, VSCode, PyCharm
Publications
-
NeurIPS 2024 Humans in Kitchens A Dataset for Multi-Person Human Motion Forecasting with Scene Context
- Julian Tanke, Oh-Hun Kwon, Felix B Mueller, Andreas Döring, Juergen Gall
-
AMFG 2023 A Gated Attention Transformer for Multi-Person Pose Tracking
- Andreas Döring, Juergen Gall
-
CVPR 2022 PoseTrack21: A Dataset for Person Search, Multi-Object Tracking and Multi-Person Pose Tracking
- Andreas Döring, Di Chen, Shanshan Zhang, Bernt Schiele and Juergen Gall
-
AAAI 2022 Keypoint Message Passing for Video-based Person Re-Identification
- Di Chen, Andreas Döring, Shanshan Zhang, Jian Yang, Juergen Gall and Bernt Schiele
-
ECCV 2020 Self-supervised Keypoint Correspondences for Multi-Person Pose Estimation and Tracking in Videos
- Andreas Döring, Umer Rafi, Bastian Leibe and Juergen Gall
-
BMVC 2018 Joint Flow: Temporal Flow Fields for Multi-Person Pose Tracking
- Andreas Döring and Juergen Gall
-
CVUI 2018 A dual-source approach for 3D human pose estimation from single images
- Umar Iqbal, Andreas Döring, Hashim Yasin, Björn Krüger, Andreas Weber and Juergen Gall