Run RKNN2 Demo on ROCK 5B

#1

Step 1. Install the latest system

You can get system images from https://github.com/radxa/debos-radxa/releases

Let’s check the kernel version:

rock@rock-5b:~$ sudo su
root@rock-5b:~# uname -r
5.10.66-14-rockchip-g05483899648a

Step 2. Get the prebuilt RKNN2 YOLOv5 demo

Download the prebuilt Yolov5 demo from https://github.com/radxa/rknpu2

root@rock-5b:~# curl https://github.com/radxa/rknpu2/releases/download/20220512/rknn_yolov5_demo_linux_20220512.tar.gz

Step 3. Run the RKNN2 YOLOv5 demo

root@rock-5b:~# tar -xvf rknn_yolov5_demo_linux_20220512.tar.gz
root@rock-5b:~# cd rknn_yolov5_demo_linux
root@rock-5b:~# export LD_LIBRARY_PATH=./lib
root@rock-5b:~# ./rknn_yolov5_demo ./model/RK3588/yolov5s-640-640.rknn ./model/bus.jpg
post process config: box_conf_threshold = 0.50, nms_threshold = 0.60
Read ./model/bus.jpg ...
img width = 640, img height = 640
Loading mode...
sdk version: 1.2.0 (1867aec5b@2022-01-14T15:16:40) driver version: 0.7.2
model input num: 1, output num: 3
  index=0, name=images, n_dims=4, dims=[1, 640, 640, 3], n_elems=1228800, size=4915200, 
fmt=NHWC, type=FP32, qnt_type=AFFINE, zp=-128, scale=0.003922
  index=0, name=output, n_dims=5, dims=[1, 3, 85, 80], n_elems=1632000, size=1632000, 
fmt=NCHW, type=INT8, qnt_type=AFFINE, zp=77, scale=0.080445
  index=1, name=371, n_dims=5, dims=[1, 3, 85, 40], n_elems=408000, size=408000, fmt=NCHW, 
type=INT8, qnt_type=AFFINE, zp=56, scale=0.080794
  index=2, name=390, n_dims=5, dims=[1, 3, 85, 20], n_elems=102000, size=102000, fmt=NCHW, 
type=INT8, qnt_type=AFFINE, zp=69, scale=0.081305
model is NHWC input fmt
model input height=640, width=640, channel=3
rga_api version 1.7.0_[1]
once run use 25.492000 ms
loadLabelName ./model/coco_80_labels_list.txt
person @ (474 250 559 523) 0.996784
person @ (112 238 208 521) 0.992814
bus @ (100 132 558 455) 0.980211
person @ (211 242 285 509) 0.976798
loop count = 10 , average run  33.581600 ms

Check the out.jpg

out
out.jpg640×640 176 KB

3 Likes
#2

@setq have you managed to get ArmNN work with LibMali as been wondering if it works better than my tries with boards with Mesa drivers due to OpenCL that seem to crash.

You could partition a model across all three of cpu, npu & gpu or run separate models on each.
Either way the Rock5b should be a AI monster capable of some pretty complex workload.

#3

I smell hardware in need of some RE work!

There’s not much there yet, just a basic wrapper library for dumping BO contents. Anyone wanting to do hardware reverse-engineering is welcome to help out.

1 Like