Wuhan University · Software Engineering · Class of 2027
Bohan Su
I am a Software Engineering undergraduate at Wuhan University. My research focuses on LLM reasoning and machine learning, particularly structured representations for computer vision and scientific problem solving. I also work on smartphone agents, embedded systems, and robotics, with experience in experiment design, system integration, and technical writing.
I am actively seeking research collaborations and RA opportunities.
Publication
ECCV 2026 · First Author
Studied person re-identification under complex appearance changes and evaluated when 3D geometry is useful across nine benchmarks.
Ongoing Research
Two first-author projects in progress
PEARL and an appearance-variation ReID survey are being developed toward AAAI and TPAMI submissions, respectively.
Engineering
Experience across several system settings
Project work includes smartphone agents, modular robots, quadruped logistics, and battery modeling, from implementation to closed-loop testing.
01 · Updates
Recent Progress
Selected research and project milestones in reverse chronological order.
First-author paper accepted to ECCV 2026
Reliability-Aware 3D Geometric Injection for Universal Person Re-identification was accepted to ECCV 2026.
PEARL development continues
The preliminary version was accepted to the WAICA Workshop. The current extension adds verifiable data curation and small-model training, with an AAAI submission as the target.
MCM 2026 Finalist
Our team combined physical modeling and data-driven methods to study smartphone battery depletion and received a Finalist Award.
National First Prize for the modular robot project
The project received a National First Prize and placed in the Top 6 at the National University IoT Design Competition.
Approach
Checking evidence and scope
My research usually starts with a specific question and evaluation protocol, followed by comparisons, ablations, and failure-case analysis. In systems projects, I pay particular attention to state transitions, recovery behavior, and end-to-end consistency.
02 · Research
Research and Ongoing Work
My current work covers two related areas: visual representation under appearance changes and scientific reasoning with structured constraints.
Reliability-Aware 3D Geometric Injection for Universal Person Re-identification
Occlusion, clothing changes, and cross-modality make appearance cues less reliable. UniGeo introduces monocular 3D body geometry as a complementary cue and uses a reliability-aware gate to limit the effect of low-confidence geometry estimates. We conducted comparisons, ablations, and perturbation-stability tests on nine benchmarks, including Market-1501 and MSMT17.
PEARL: Verifiable Curation and Small-Model Learning for Scientific Reasoning Graph Extraction
A scientific reasoning graph must be semantically meaningful while also satisfying structural, compilation, and source-alignment constraints. PEARL separates model-generated candidates from code-based checks. The preliminary version validated this workflow; the full version uses curated graphs to train smaller models and is being extended toward an AAAI submission.
Appearance Variations in Person Re-Identification: A Survey
Existing ReID studies often discuss pose, occlusion, clothing changes, cross-modality, and long-term deployment separately, although their boundaries are not always clear. This survey reorganizes the field around short-term perturbations, long-term appearance changes, and compound variations, and reviews methods and evaluation through the distinction between appearance robustness and appearance generalization. Model fine-tuning and testing are also in progress.
03 · Systems
Selected Projects
These projects span smartphone agents, embedded communication, robotics, and mathematical modeling, with work across system design, algorithm development, and engineering implementation.
A4R: Guided Smartphone AI Assistant
I co-designed a clarify-execute-verify interaction workflow and built more than 150 real interactions across six domains and 16 applications. On China Unicom tasks, the system achieved a 64.68% completion rate and a 72.33% effective-guidance rate.
Modular Robot System
I was responsible for module coordination, bus arbitration, hot-plug discovery, and bus merging, allowing the controller to rediscover topology and restore communication after modules were added or reordered. The system reached a 98.5% communication success rate and improved communication efficiency by about 40% over the original design.
SmartPhone Battery: The Moment of Depletion
Our team modeled screen, computation, and communication loads in a shared physical framework, then cross-checked an ECM-EKF solver with a physics-informed neural network. Across eight data sources and more than 120,000 time steps, SOC RMSE was 1.24% and overall TTE MAPE was 8.7%.
Xiaomi Quadruped Logistics System
I contributed to system design, algorithm debugging, workflow optimization, and on-site testing. My work focused on state transitions and action timing between delivery stages, reducing the downstream effects of recognition failures and path deviations.
04 · Background
Education, Recognition, and Service
Selected academic, competition, and university experience.
Education
Wuhan University
Bachelor's student in Software Engineering, Sep 2023 - Jun 2027 (expected)
- GPA: 3.75 / 4.0
- Coursework includes computer vision, algorithms, networks, operating systems, databases, and software engineering
Selected Recognition
Competitions and Scholarships
- National University IoT Design Competition: National First Prize, Top 6
- China Youth Sci-Tech Innovation Challenge: National Second Prize
- Xiaomi Cup: National Third Prize; MCM 2026: Finalist
- Zheng Geru First-Class Scholarship: first in the grade-level defense
- Wuhan University First-Class Scholarship and Merit Student, twice each
University Service
Student Activities
I served as Vice Minister in the Publicity Department of the Student Party Branch Secretaries Joint Council, contributing to article coordination and publishing-process improvements across more than 100 articles. I also worked on quality monitoring and issue follow-up for Wuhan University's Second Classroom platform.
05 · Contact
Actively Seeking Research Collaborations and RA Opportunities
If you are working on computer vision, structured reasoning, or multi-agent systems and think my experience may fit your project, please feel free to contact me by email.