GPU Programming In Computer Vision: Practical Course At TUM, Summer Semester 2018.pdf

a practical course on GPU programming in computer vision offered by the Computer Vision Group at Technische Universität München during the Summer Semester of 2018. The course is taught by Björn Häfner, Robert Maier, and David Schubert.

Key learning objectives include an introduction to parallel computing on GPUs, the NVIDIA CUDA Framework, and how to parallelize basic computer vision algorithms using CUDA/C++. Students will gain practical project experience and develop teamwork and presentation skills.

Important dates are provided: a preliminary meeting on February 7th, registration from February 9th to 14th, and a deadline for sending CVs by February 13th. Matching results were announced on February 21st, with only assigned students allowed to attend.

The course is structured as a 4–5 week block during the semester break (beginning of September - mid-October), including one week of lectures and exercises followed by a 3–4 week project phase. Students work in groups of three, each equipped with a GTX 750 GPU. Each group receives an advisor.

Prerequisites include good knowledge of C/C++ and basic mathematics (calculus/analysis and linear algebra). The first week involves lectures on CUDA and math, along with hands-on programming exercises. Subsequent weeks focus on project work, where students propose their own ideas related to computer vision, image processing, or machine learning, culminating in a demo day presentation.

Evaluation criteria include successful completion of exercises (with a 0.3 bonus), expertise gained in CUDA/parallel programming, quality and completion of the final project, and performance during presentations. Regular attendance is mandatory, with medical attest required for absences. The course emphasizes the motivation behind GPU programming by comparing CPU and GPU capabilities, highlighting the significant increase in cores available on GPUs (e.g., 4-32 cores vs. 3072 cores). Examples from previous student projects are also mentioned.