Research Collaboration Call
Collaborate on Human-Centric Vision Projects
I am open to research collaborations on human-centric computer vision, with emphasis on spine and whole-body motion understanding.
This page is for project-based collaboration (co-design, benchmarking, data, modeling, and papers), not team hiring or recruiting.
3
Active Areas
12
Topics Supervised
1
Theses
4
Semester Cycles
Open Collaboration Areas
Current areas where external collaboration is welcome.
Extend benchmarking protocols, add stronger reference baselines, and improve reproducible evaluation for spine-aware motion understanding.
- Experience with 2D/3D pose estimation or biomechanics-informed modeling
- Strong experimental design, ablation studies, and reproducibility practices
Study continual adaptation of pose models under realistic domain shift and evolving task definitions, together with sparsity and efficiency constraints for sustainable deployment.
- Continual learning, adaptation under memory constraints, and robust evaluation
- Model sparsification, optimization, and deployment-aware benchmarking
Develop promptable pose estimation and compact multimodal models for human-centric tasks, combining category-agnostic keypoint reasoning with efficient vision-language adaptation.
- PyTorch and modern detection/pose codebases (for example OpenMMLab)
- Experience with foundation models, promptable learning, distillation, or zero-shot transfer
Collaboration Modes
- Co-authored research studies: Joint problem definition, method design, experimentation, and writing for workshops or conferences.
- Benchmark and reproducibility partnerships: Re-running baselines, stress-testing methods, and improving evaluation protocols across datasets.
- Data and annotation collaborations: Building or refining datasets, quality control pipelines, and task definitions for new labels.
- Applied transfer collaborations: Adapting research code to a partner's domain while keeping scientific rigor and reproducibility.
Typical Workflow
- Align on research question, deliverable, and timeline.
- Define protocol, baselines, and evaluation criteria up front.
- Run implementation and experiments with transparent checkpoints.
- Package outcomes for reproducibility and publication readiness.
Academic Supervision Track Record
Since 2023: 1 thesis topics, 2 guided research topics, and 4 semester cycles of projects and seminars.
Theses
- Thesis Classroom-Inspired Learning for Deep Neural Networks - Shalini Sarode (Oct 2024)
Guided Research
- Guided Research Enhancing Zero-Shot Classification with CLIP-Driven Optimal Natural Language Description Selection - Nosheen Nazir (Nov 2024)
- Guided Research Monocular 3D Human Pose Estimation Using Depth Augmentation - Jeremias Krauss (Oct 2024)
Previous Semesters
Winter Semester 2024/25
- Project Human Image Generation through Multimodal Diffusion Models - Muhammad Usama
Summer Semester 2024
- Seminar Advances in Gait Recognition - Shino Sam
- Project Gait Recognition using OpenGait - Aakarsh Goel
- Seminar Controllable Style Transfer for Pose-Guided Human Image Generation Using Diffusion - Muhammad Usama
Winter Semester 2023/24
- Project Unified 2D Human Pose Estimation - Dhavalkumar Limbachiya
- Project Active Learning for Optimal Data Labeling - Nikita Khutorni
Summer Semester 2023
- Seminar Deep Neural Networks for Vision Tasks on Mobile Devices - Aditya Ranjan Dash
- Seminar Automating Design of Efficient Models with Neural Architecture Search - Sebastian Igel
- Project Optimizing Human Pose Estimation through Knowledge Distillation - Sebastian Igel, Shalini Sarode