Closed Mercotui closed 5 years ago
A first version of this is implemented in f19e6386e4052507a7cfb6362cfdb7126efbcff1 It uses OpenGL-ES 3.0 to do downscaling, blurring and segmentation on the GPU. I want to test this on android before claiming victory, but this looks pretty promising.
Merged it into devel :D
The current CPU based algorithm for vision processing and feature extraction is a significant bottleneck. Especially the android app performance, running on low power CPUs suffers greatly.
Although the original intention was to implement the algorithm as an OpenGL-ES 3.1 compute shader, neither of the developers own phones that support this version of OpenGL-ES. Therefore, most likely OpenCL will be used.
Originally mentioned at #3.