易百纳rv1126 a201开发板跑rknn yolov5
一开始我以为跑yolov5啊, 那必须上debian/ubuntu啊, 不然怎么装python, pytorch?buildroot没有apt, 于是花了大量时间想办法在a201的板子上跑debian, 等我debian跑上, python安排上, rknn-lite按照firefly的方法装好
https://dev.t-firefly.com/thread-104204-1-1.html
再去找rknn的示例程序的时候, 发现原来有c的版本.
https://github.com/airockchip/rknn_model_zoo/tree/main/models/vision/object_detection/yolov5-pytorch/RKNN_C_demo/RKNN_toolkit_1/rknn_yolov5_demo
默认易百纳的出厂固件就有rknn的lib
[root@RV1126_RV1109:/etc]# echo $LD_LIBRARY_PATH
:/oem/usrlibs:/oem/usr/lib:/oem/lib
[root@RV1126_RV1109:/etc]# cd /oem/usr/lib
[root@RV1126_RV1109:/oem/usr/lib]# ls
face_attribute.data libeasymedia.so.1 librockchip_mpp.so.0
face_detection_v3.data libeasymedia.so.1.0.1 librockchip_mpp.so.1
face_detection_v3_large.data libgdbus.so librockchip_vpu.so
face_landmark5.data libispclient.so librockchip_vpu.so.0
face_landmarks106.data libmal.so librockchip_vpu.so.1
face_landmarks68.data librkaiq.so librockface.so
face_recognition.data librkdb.so librockx.so
libCallFunIpc.so librkisp_api.so object_detection.data
libIPCProtocol.so librknn_api.so pose_body_v2.data
libeasymedia.so librockchip_mpp.so pose_finger.data
实现方法， 第一步， 用rkmedia产生一个jpeg图片，这个就简单了， 直接改rkmedia的example里面的示例代码，但是原本是用的CMakelists编译的， 效率太低， 直接改Makefile
hide := @
ECHO := echo
GCC := /home/marc/rv_sdk/prebuilts/gcc/linux-x86/arm/gcc-arm-8.3-2019.03-x86_64-arm-linux-gnueabihf/bin/arm-linux-gnueabihf-gcc
SYSROOT = /home/marc/rv_sdk/buildroot/output/rockchip_rv1126_rv1109/host/arm-buildroot-linux-gnueabihf/sysroot
CFLAGS := -I../include/rkmedia \
-I../include/rkaiq/common \
-I../include/rkaiq/xcore \
-I../include/rkaiq/uAPI \
-I../include/rkaiq/algos \
-I../include/rkaiq/iq_parser\
#-I/home/marc/rv_sdk/buildroot/output/rockchip_rv1126_rv1109/build/camera_engine_rkaiq-1.0/include/uAPI/\
#-I/home/marc/rv_sdk/buildroot/output/rockchip_rv1126_rv1109/build/camera_engine_rkaiq-1.0/include/xcore/\
#-I/home/marc/rv_sdk/buildroot/output/rockchip_rv1126_rv1109/build/camera_engine_rkaiq-1.0/include/iq_parser/\
LIB_FILES := -L$(SYSROOT)/usr/lib
LD_FLAGS := -lpthread -leasymedia -ldrm -lrockchip_mpp \
-lavformat -lavcodec -lswresample -lavutil \
-lasound -lv4l2 -lv4lconvert -lrga \
-lRKAP_ANR -lRKAP_Common -lRKAP_3A \
-lmd_share -lrkaiq -lod_share
CFLAGS += -DRKAIQ
SAMPLE_COMMON := common/sample_common_isp.c
all:
$(GCC) jpeg_cap.c $(SAMPLE_COMMON) $(LIB_FILES) $(LD_FLAGS) $(CFLAGS) -o build/jpeg_capture --sysroot=$(SYSROOT)
# $(GCC) my_vi.c $(SAMPLE_COMMON) $(LIB_FILES) $(LD_FLAGS) $(CFLAGS) -o build/my_vi --sysroot=$(SYSROOT)
$(hide)$(ECHO) "Build Done ..."
就是改下GCC的目录跟加入sysroot目录, 不然会提示找不到库
通过make编译一下这个获取图片并保存的app:
#include 上面的代码执行起来, 会不停的删除之前的图片,抓取jpeg编码后的数据,写入新的同名图片, 保存在一个nginx的www目录下面, 方便我们后期用一个自动刷新的网站来显示使用yolov5做predict之前的图片.
这样我们就获得了一张从摄像头获取的jpeg图片. 且是不断刷新的.
然后使用前面说的rknn的示例代码, 进行推测:
注意几点:
- 使用模型要复制到app执行的目录下面, 名字是yolov5s_relu_rv1109_rv1126_out_opt.rknn
- COCO的标签也需要复制过去
- 原本执行的参数, 我直接写死了, 省得每次都要打参数进去.
- 输出路径同为nginx的www目录, 不过我的nginx的是跑在我开发环境的ubuntu上面的, 所以目录不能搞乱了, 其实可以考虑把nginx跑在开发板上, 只是需要拿nginx源码编译一下就行.
具体代码main.cc如下:
/*-------------------------------------------
Includes
-------------------------------------------*/
#include [fp/u8] [single_img/multi_imgs] \n" , argv[0]);
// return -1;
}
// model_name = (char *) argv[1];
// char *input_type = argv[3];
// char *input_path = argv[4];
printf("Post process with u8\n");
printf("Test with single img\n");
/* Create the neural network */
printf("Loading model...\n");
int model_data_size = 0;
char *modelName = "/mnt/my_stuff/yolov5s_relu_rv1109_rv1126_out_opt.rknn";
unsigned char *model_data = load_model(modelName, &model_data_size);
printf("model_name:%s model_data_size:%d\n", modelName, model_data_size);
ret = rknn_init(&ctx, model_data, model_data_size, 0);
if (ret < 0) {
printf("rknn_init error ret=%d\n", ret);
return -1;
}
printf("Query SDK version...\n");
/* Query sdk version */
rknn_sdk_version version;
ret = rknn_query(ctx, RKNN_QUERY_SDK_VERSION, &version, sizeof(rknn_sdk_version));
if (ret < 0) {
printf("rknn_init error ret=%d\n", ret);
return -1;
}
printf("sdk version: %s driver version: %s\n", version.api_version, version.drv_version);
/* Get input,output attr */
rknn_input_output_num io_num;
ret = rknn_query(ctx, RKNN_QUERY_IN_OUT_NUM, &io_num, sizeof(io_num));
if (ret < 0) {
printf("rknn_init error ret=%d\n", ret);
return -1;
}
printf("model input num: %d, output num: %d\n", io_num.n_input, io_num.n_output);
rknn_tensor_attr input_attrs[io_num.n_input];
memset(input_attrs, 0, sizeof(input_attrs));
for (int i = 0; i < io_num.n_input; i++) {
input_attrs[i].index = i;
ret = rknn_query(ctx, RKNN_QUERY_INPUT_ATTR, &(input_attrs[i]), sizeof(rknn_tensor_attr));
if (ret < 0) {
printf("rknn_init error ret=%d\n", ret);
return -1;
}
printRKNNTensor(&(input_attrs[i]));
}
rknn_tensor_attr output_attrs[io_num.n_output];
memset(output_attrs, 0, sizeof(output_attrs));
for (int i = 0; i < io_num.n_output; i++) {
output_attrs[i].index = i;
ret = rknn_query(ctx, RKNN_QUERY_OUTPUT_ATTR, &(output_attrs[i]), sizeof(rknn_tensor_attr));
printRKNNTensor(&(output_attrs[i]));
}
int input_channel = 3;
int input_width = 0;
int input_height = 0;
if (input_attrs[0].fmt == RKNN_TENSOR_NCHW) {
printf("model is NCHW input fmt\n");
input_width = input_attrs[0].dims[0];
input_height = input_attrs[0].dims[1];
} else {
printf("model is NHWC input fmt\n");
input_width = input_attrs[0].dims[1];
input_height = input_attrs[0].dims[2];
}
printf("model input height=%d, width=%d, channel=%d\n", input_height, input_width, input_channel);
/* Init input tensor */
rknn_input inputs[1];
memset(inputs, 0, sizeof(inputs));
inputs[0].index = 0;
inputs[0].type = RKNN_TENSOR_UINT8;
inputs[0].size = input_width * input_height * input_channel;
inputs[0].fmt = RKNN_TENSOR_NHWC;
inputs[0].pass_through = 0;
/* Init output tensor */
rknn_output outputs[io_num.n_output];
memset(outputs, 0, sizeof(outputs));
for (int i = 0; i < io_num.n_output; i++) {
outputs[i].want_float = 0;
}
void *resize_buf = malloc(input_height * input_width * input_channel);
// unsigned char *input_data = NULL;
/* Single img input */
printf("Single img input\n");
while (1) {
predict(output_attrs, outputs, inputs, &io_num,
input_width, input_height, input_channel, input_attrs, resize_buf);
usleep(10000);
}
// release
ret = rknn_destroy(ctx);
RGA_deinit(&rga_ctx);
if (model_data) {
free(model_data);
}
if (resize_buf) {
free(resize_buf);
}
return 0;
}
使用自带的build.sh脚本编译安装就行.
接着 把两个应用跟nginx都跑起来, 就可以看到整个效果了.
我特么红外又忘了关了。。。
最后这个自动刷新的网站的html脚本如下:
DOCTYPE html>
<html>
<head>
<title>Welcome to nginx!title>
<style>
body {
width: 35em;
margin: 0 auto;
font-family: Tahoma, Verdana, Arial, sans-serif;
}
style>
head>
<body>
<center>
<b>Marc's rknn_yolov5 play groundb>br>
<img src="rknn_before.jpeg" />
<img src="rknn_after.bmp" />
center>
body>
<script language="Javascript">
function myrefresh() {
window.location.reload();
}
setTimeout('myrefresh ()', 1000); // 指定 1 秒刷新一次（1401）
script>
html>