This is my undergraduate research project. For two semesters and a summer I:
- Designed and developed a parallel implementation of the Probabilistic Hypergraph Matching Algorithm on Nvidia Tesla GPUs using CUDA.
- Achieved 10x speed up compared to CPU serial implementation
- Used Thrust parallel template library to simplify programming tasks
- Made image matching demo and used OpenSURF for feature detection.
The research result was presented at GPU Technology Conference 2015. In the process of this project I learned and applied parallel patterns, visual profiling, throughput-oriented optimization and load-balancing techniques.
The project source code can be found here on Github.