GPUs are particularly powerful in compute-intensive image processing and may serve in many environments requiring realtime operation, such as video surveillance, traffic analysis or in-vehicle safety systems. Many image progessing algorithms could be easily implemented using existing building blocks, such as OpenCV. However, additional performance and flexibility requires deeper knowledge of GPU internals and programming techniques.
Applied Parallel Computing LLC offers a specialized 3-day course on Image Processing with CUDA. The first day is dedicated to the basics of GPU architecture and CUDA programming. The second and the third days of training are dedicated to intensive guided CUDA practice in implementing different types of image filters. Where appliciable, CUDA implementations are compared to OpenCV.