vpu

Nanopi M4 RK3399 base minimal image for development (mali fbdev / gbm) - Camera support

YOLO multi-threaded and hardware-accelerated inference framework based on RKNN

Library for modelling performance costs of different Neural Network workloads on NPU devices

Website for Eclipse ioFog, a distributed Edge Compute Network (ECN) platform

Kodi master (GBM) with hw video decoding for Rockchip on Armbian, as containers & .debs. Two flavors: rkmpp (vendor kernel; ffmpeg-rockchip MPP+RGA via boogie/nyanmisaka/joshua/amazingfate/reardonia/chewitt magic) and v4l2requests (mainline kernel; FFmpeg V4L2 Request API via LibreELEC/Kwiboo/jernejsk/Collabora/chewitt magic).

The Rockchip VPU driver module code for Linux, taken from rockchip-forwardports. I'm still hesitating between rockchip-vcodec and literate-octo-guacamole though...

Rockchip low latency video decoding

Zero-copy four-camera real-time panorama stitching on Rockchip RK3588: VPU(MPP) hardware decode -> Mali GPU OpenCL warp -> RGA blend, DMA-BUF zero-copy

Single Shot MultiBox Detector deployed on a OAK-D Lite cam via DepthAI

A ROS package for offloading inference to the Intel Movidius VPU Neural Compute Stick. Also, Evaluation of the Intel Movidius VPU Neural Compute Stick for Real-Time Inference.

This repository contains detailed notes of all chapters and all three projects completed in Intel-Edge-AI NanoDegree.

RK3588 development tutorial: OpenCV, V4L2, MPP decode+encode, Mali GPU OpenCL, NPU RKNN - six runnable examples

Simple video player for NanoPi M3

RK3588 panorama stitching evolution Gen1: pure CPU image warping + stitching with NPU crack detection (YOLOv8), self-contained CMake build

Standalone USB camera H.264 streaming on Rockchip RK3588: V4L2 DMABUF MJPG -> MPP decode -> MPP H.264 encode -> TCP push :5001, single-client replacement, zero-copy

The main focus area of this project is to propose a possible hardware solution for each scenario. Build out the application and test its performance on the DevCloud using multiple hardware types. Compare the performance to see which hardware performed best.

working on picorv-based accelerator...

Full-pipeline integration on Rockchip RK3588: panorama stitching DMA-BUF -> tri-core NPU detection -> plugin output (TCP H.264 streaming + UDP stats), zero-copy end-to-end

RK3588 panorama stitching evolution Gen2: Mali GPU (OpenCL) image warp worker, DMA-BUF zero-copy