Acceleration of Deep Learning Inference on Raspberry Pi's VideoCore GPU

Explore the acceleration of deep learning inference on Raspberry Pi's VideoCore GPU in this 26-minute conference talk from tinyML Asia 2020. Discover how Koichi NAKAMURA from Idein Inc. leverages the underutilized VideoCore IV/VI GPU to achieve significant speedups in machine learning models without compromising accuracy. Learn about the development of specialized device programming tools, libraries, math kernels, and an optimizing graph compiler for VideoCore GPGPU usage. Gain insights into the architectural features of VideoCore, acceleration techniques, and additional research on ARM CPUs, Intel GPUs, and FPGAs. Delve into topics such as hardware benchmarks, software tools, Video 4 Architecture, algorithm selection, HWC layout, conversion corners, and strategies to remove I/O control overhead and address CPU bottlenecks.

Introduction
About Koichi
Why Raspberry Pi
Demos
Hardware
Benchmarks
Software Tools
Intel Software
Video 4 Architecture
Algorithm Selection
HWC Layout
Conversion Corners
Remove IO Control Overhead
CPU bottleneck
Activecast
Conclusion