forked from xuos/xiuos
update knowing framework and application from Tian_Chunyu
it is perfect
This commit is contained in:
xuedongliang
@@ -1,14 +1,3 @@
|
||||
menu "connection app"
|
||||
menuconfig APPLICATION_CONNECTION
|
||||
bool "Using connection apps"
|
||||
default n
|
||||
|
||||
menuconfig CONNECTION_COMMUNICATION_ZIGBEE
|
||||
bool "enable zigbee demo"
|
||||
default n
|
||||
select CONFIG_CONNECTION_COMMUNICATION_ZIGBEE
|
||||
if CONNECTION_COMMUNICATION_ZIGBEE
|
||||
source "$KERNEL_DIR/framework/connection/Adapter/zigbee/Kconfig"
|
||||
endif
|
||||
|
||||
endmenu
|
||||
|
||||
@@ -1,7 +1,4 @@
|
||||
menu "knowing app"
|
||||
|
||||
menuconfig APPLICATION_KNOWING
|
||||
bool "Using knowing apps"
|
||||
default n
|
||||
|
||||
source "$APP_DIR/Applications/knowing_app/mnist/Kconfig"
|
||||
source "$APP_DIR/Applications/knowing_app/face_detect/Kconfig"
|
||||
endmenu
|
||||
|
||||
@@ -0,0 +1,14 @@
|
||||
import os
|
||||
Import('RTT_ROOT')
|
||||
from building import *
|
||||
|
||||
cwd = GetCurrentDir()
|
||||
objs = []
|
||||
list = os.listdir(cwd)
|
||||
|
||||
for d in list:
|
||||
path = os.path.join(cwd, d)
|
||||
if os.path.isfile(os.path.join(path, 'SConscript')):
|
||||
objs = objs + SConscript(os.path.join(path, 'SConscript'))
|
||||
|
||||
Return('objs')
|
||||
@@ -0,0 +1,7 @@
|
||||
config FACE_DETECT
|
||||
bool "enable apps/face detect"
|
||||
depends on BOARD_K210_EVB
|
||||
depends on DRV_USING_OV2640
|
||||
depends on USING_KPU_POSTPROCESSING
|
||||
depends on USING_YOLOV2
|
||||
default n
|
||||
@@ -0,0 +1,9 @@
|
||||
from building import *
|
||||
|
||||
cwd = GetCurrentDir()
|
||||
src = Glob('*.c') + Glob('*.cpp')
|
||||
CPPPATH = [cwd]
|
||||
|
||||
group = DefineGroup('Applications', src, depend = ['FACE_DETECT'], LOCAL_CPPPATH = CPPPATH)
|
||||
|
||||
Return('group')
|
||||
@@ -0,0 +1,253 @@
|
||||
#include <transform.h>
|
||||
#include"region_layer.h"
|
||||
#define SHOW_RGB_BUF_SIZE (320*240*2)
|
||||
#define AI_KPU_RGB_BUF_SIZE (320*240*3)
|
||||
#define KMODEL_SIZE (388776) //face model size
|
||||
#define ANCHOR_NUM 5
|
||||
#define KPUIMAGEWIDTH (320)
|
||||
#define KPUIMAGEHEIGHT (240)
|
||||
|
||||
|
||||
static float anchor[ANCHOR_NUM * 2] = {1.889,2.5245, 2.9465,3.94056, 3.99987,5.3658, 5.155437,6.92275, 6.718375,9.01025};
|
||||
|
||||
|
||||
#define THREAD_PRIORITY_FACE_D (11)
|
||||
static pthread_t facetid = 0;
|
||||
static void* thread_face_detcet_entry(void *parameter);
|
||||
static int g_fd = 0;
|
||||
static int kmodel_fd = 0;
|
||||
static int if_exit = 0;
|
||||
static unsigned char * showbuffer = NULL ;
|
||||
static unsigned char * kpurgbbuffer = NULL ;
|
||||
|
||||
static _ioctl_shoot_para shoot_para_t = {0};
|
||||
unsigned char * model_data = NULL; //kpu data load memory
|
||||
unsigned char *model_data_align = NULL;
|
||||
|
||||
kpu_model_context_t face_detect_task;
|
||||
static region_layer_t face_detect_rl;
|
||||
static obj_info_t face_detect_info;
|
||||
volatile uint32_t g_ai_done_flag;
|
||||
|
||||
static void ai_done(void *ctx)
|
||||
{
|
||||
g_ai_done_flag = 1;
|
||||
}
|
||||
|
||||
|
||||
|
||||
void face_detect()
|
||||
{
|
||||
int ret = 0;
|
||||
int result = 0;
|
||||
int size = 0;
|
||||
g_fd = open("/dev/ov2640",O_RDONLY);
|
||||
if(g_fd < 0)
|
||||
{
|
||||
printf("open ov2640 fail !!");
|
||||
return;
|
||||
}
|
||||
showbuffer = (unsigned char*)malloc(SHOW_RGB_BUF_SIZE);
|
||||
if(NULL ==showbuffer)
|
||||
{
|
||||
close(g_fd);
|
||||
printf("showbuffer apply memory fail !!");
|
||||
return ;
|
||||
}
|
||||
kpurgbbuffer = (unsigned char*)malloc(AI_KPU_RGB_BUF_SIZE);
|
||||
if(NULL ==kpurgbbuffer)
|
||||
{
|
||||
close(g_fd);
|
||||
free(showbuffer);
|
||||
printf("kpurgbbuffer apply memory fail !!");
|
||||
return ;
|
||||
}
|
||||
model_data = (unsigned char *)malloc(KMODEL_SIZE + 255);
|
||||
if(NULL ==model_data)
|
||||
{
|
||||
free(showbuffer);
|
||||
free(kpurgbbuffer);
|
||||
close(g_fd);
|
||||
printf("model_data apply memory fail !!");
|
||||
return ;
|
||||
}
|
||||
memset(model_data,0,KMODEL_SIZE + 255);
|
||||
memset(showbuffer,0,SHOW_RGB_BUF_SIZE);
|
||||
memset(kpurgbbuffer,0,AI_KPU_RGB_BUF_SIZE);
|
||||
shoot_para_t.pdata = (unsigned int *)(showbuffer);
|
||||
shoot_para_t.length = SHOW_RGB_BUF_SIZE;
|
||||
/*
|
||||
load memory
|
||||
*/
|
||||
kmodel_fd = open("/kmodel/detect.kmodel",O_RDONLY);
|
||||
if(kmodel_fd <0)
|
||||
{
|
||||
printf("open kmodel fail");
|
||||
close(g_fd);
|
||||
free(showbuffer);
|
||||
free(kpurgbbuffer);
|
||||
free(model_data);
|
||||
return;
|
||||
}
|
||||
else
|
||||
{
|
||||
size = read(kmodel_fd, model_data, KMODEL_SIZE);
|
||||
if(size != KMODEL_SIZE)
|
||||
{
|
||||
printf("read kmodel error size %d\n",size);
|
||||
close(g_fd);
|
||||
close(kmodel_fd);
|
||||
free(showbuffer);
|
||||
free(kpurgbbuffer);
|
||||
free(model_data);
|
||||
return;
|
||||
|
||||
}
|
||||
else
|
||||
{
|
||||
printf("read kmodel success \n");
|
||||
}
|
||||
|
||||
}
|
||||
unsigned char *model_data_align = (unsigned char *)(((unsigned int)model_data+255)&(~255));
|
||||
dvp_set_ai_addr((uint32_t)kpurgbbuffer, (uint32_t)(kpurgbbuffer + 320 * 240), (uint32_t)(kpurgbbuffer + 320 * 240 * 2));
|
||||
if (kpu_load_kmodel(&face_detect_task, model_data_align) != 0)
|
||||
{
|
||||
printf("\nmodel init error\n");
|
||||
close(g_fd);
|
||||
close(kmodel_fd);
|
||||
free(showbuffer);
|
||||
free(kpurgbbuffer);
|
||||
free(model_data);
|
||||
return;
|
||||
}
|
||||
face_detect_rl.anchor_number = ANCHOR_NUM;
|
||||
face_detect_rl.anchor = anchor;
|
||||
face_detect_rl.threshold = 0.7;
|
||||
face_detect_rl.nms_value = 0.3;
|
||||
result = region_layer_init(&face_detect_rl, 20, 15, 30, KPUIMAGEWIDTH, KPUIMAGEHEIGHT);
|
||||
printf("region_layer_init result %d \n\r",result);
|
||||
size_t stack_size = 32*1024;
|
||||
pthread_attr_t attr; /* 线程属性 */
|
||||
struct sched_param prio; /* 线程优先级 */
|
||||
prio.sched_priority = 8; /* 优先级设置为 8 */
|
||||
pthread_attr_init(&attr); /* 先使用默认值初始化属性 */
|
||||
pthread_attr_setschedparam(&attr,&prio); /* 修改属性对应的优先级 */
|
||||
pthread_attr_setstacksize(&attr, stack_size);
|
||||
|
||||
/* 创建线程 1, 属性为 attr，入口函数是 thread_entry，入口函数参数是 1 */
|
||||
result = pthread_create(&facetid,&attr,thread_face_detcet_entry,NULL);
|
||||
if (0 == result)
|
||||
{
|
||||
printf("thread_face_detcet_entry successfully!\n");
|
||||
}
|
||||
else
|
||||
{
|
||||
printf("thread_face_detcet_entry failed! error code is %d\n",result);
|
||||
close(g_fd);
|
||||
}
|
||||
}
|
||||
#ifdef __RT_THREAD_H__
|
||||
MSH_CMD_EXPORT(face_detect,face detect task);
|
||||
#endif
|
||||
static void* thread_face_detcet_entry(void *parameter)
|
||||
{
|
||||
extern void lcd_draw_picture(uint16_t x1, uint16_t y1, uint16_t width, uint16_t height, uint32_t *ptr);
|
||||
printf("thread_face_detcet_entry start!\n");
|
||||
int ret = 0;
|
||||
//sysctl_enable_irq();
|
||||
while(1)
|
||||
{
|
||||
//memset(showbuffer,0,320*240*2);
|
||||
g_ai_done_flag = 0;
|
||||
ret = ioctl(g_fd,IOCTRL_CAMERA_START_SHOT,&shoot_para_t);
|
||||
if(RT_ERROR == ret)
|
||||
{
|
||||
printf("ov2640 can't wait event flag");
|
||||
rt_free(showbuffer);
|
||||
close(g_fd);
|
||||
pthread_exit(NULL);
|
||||
return NULL;
|
||||
}
|
||||
kpu_run_kmodel(&face_detect_task, kpurgbbuffer, DMAC_CHANNEL5, ai_done, NULL);
|
||||
while(!g_ai_done_flag);
|
||||
float *output;
|
||||
size_t output_size;
|
||||
kpu_get_output(&face_detect_task, 0, (uint8_t **)&output, &output_size);
|
||||
face_detect_rl.input = output;
|
||||
region_layer_run(&face_detect_rl, &face_detect_info);
|
||||
/* display result */
|
||||
#ifdef BSP_USING_LCD
|
||||
for (int face_cnt = 0; face_cnt < face_detect_info.obj_number; face_cnt++)
|
||||
{
|
||||
draw_edge((uint32_t *)showbuffer, &face_detect_info, face_cnt, 0xF800);
|
||||
}
|
||||
lcd_draw_picture(0, 0, 320, 240, (unsigned int*)showbuffer);
|
||||
#endif
|
||||
usleep(1);
|
||||
if(1 == if_exit)
|
||||
{
|
||||
if_exit = 0;
|
||||
printf("thread_face_detcet_entry exit");
|
||||
pthread_exit(NULL);
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
void face_detect_delete()
|
||||
{
|
||||
if(showbuffer != NULL)
|
||||
{
|
||||
int ret = 0;
|
||||
close(g_fd);
|
||||
close(kmodel_fd);
|
||||
free(showbuffer);
|
||||
free(kpurgbbuffer);
|
||||
free(model_data);
|
||||
printf("face detect task cancel!!! ret %d ",ret);
|
||||
if_exit = 1;
|
||||
}
|
||||
|
||||
}
|
||||
#ifdef __RT_THREAD_H__
|
||||
MSH_CMD_EXPORT(face_detect_delete,face detect task delete);
|
||||
#endif
|
||||
void kmodel_load(unsigned char * model_data)
|
||||
{
|
||||
int kmodel_fd = 0;
|
||||
int size = 0;
|
||||
kmodel_fd = open("/kmodel/detect.kmodel",O_RDONLY);
|
||||
|
||||
model_data = (unsigned char *)malloc(KMODEL_SIZE + 255);
|
||||
if(NULL ==model_data)
|
||||
{
|
||||
printf("model_data apply memory fail !!");
|
||||
return ;
|
||||
}
|
||||
memset(model_data,0,KMODEL_SIZE + 255);
|
||||
|
||||
if (kmodel_fd>= 0)
|
||||
{
|
||||
size = read(kmodel_fd, model_data, KMODEL_SIZE);
|
||||
if(size != KMODEL_SIZE)
|
||||
{
|
||||
printf("read kmodel error size %d\n",size);
|
||||
|
||||
}
|
||||
else
|
||||
{
|
||||
printf("read kmodel success");
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
free(model_data);
|
||||
printf("open kmodel fail");
|
||||
}
|
||||
|
||||
}
|
||||
#ifdef __RT_THREAD_H__
|
||||
MSH_CMD_EXPORT(kmodel_load,kmodel load memory);
|
||||
#endif
|
||||
|
||||
@@ -0,0 +1,2 @@
|
||||
*.h5
|
||||
*.tflite
|
||||
Binary file not shown.
|
After Width: | Height: | Size: 32 KiB |
@@ -0,0 +1,4 @@
|
||||
config APP_MNIST
|
||||
bool "enable apps/mnist"
|
||||
depends on USING_TENSORFLOWLITEMICRO
|
||||
default n
|
||||
@@ -0,0 +1,23 @@
|
||||
# MNIST 说明
|
||||
|
||||
要使用本例程，MCU RAM必须至少500K左右，所以本例程目前在K210上面验证过，stm32f407 目前在rtt上原则上只能采取dlmodule加载的方式。
|
||||
|
||||
|
||||
|
||||
## 使用
|
||||
|
||||
tools/mnist-train.py 训练生成 mnist 模型。
|
||||
|
||||
tools/mnist-inference.py 使用 mnist 模型进行推理。
|
||||
|
||||
tools/mnist-c-model.py 将 mnist 模型转换成 C 的数组保存在 model.h 中。
|
||||
|
||||
tools/mnist-c-digit.py 将 mnist 数据集中的某个数字转成数组保存在 digit.h 中。
|
||||
|
||||
## 参考资料
|
||||
|
||||
https://tensorflow.google.cn/lite/performance/post_training_quantization
|
||||
|
||||
https://tensorflow.google.cn/lite/performance/post_training_integer_quant
|
||||
|
||||
https://colab.research.google.com/github/tensorflow/tensorflow/blob/master/tensorflow/lite/micro/examples/hello_world/train/train_hello_world_model.ipynb
|
||||
@@ -0,0 +1,9 @@
|
||||
from building import *
|
||||
|
||||
cwd = GetCurrentDir()
|
||||
src = Glob('*.c') + Glob('*.cpp')
|
||||
CPPPATH = [cwd]
|
||||
|
||||
group = DefineGroup('Applications', src, depend = ['APP_MNIST'], LOCAL_CPPPATH = CPPPATH)
|
||||
|
||||
Return('group')
|
||||
@@ -0,0 +1,31 @@
|
||||
const float mnist_digit[] = {
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.33, 0.73, 0.62, 0.59, 0.24, 0.14, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.87, 1.00, 1.00, 1.00, 1.00, 0.95, 0.78, 0.78, 0.78, 0.78, 0.78, 0.78, 0.78, 0.78, 0.67, 0.20, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.26, 0.45, 0.28, 0.45, 0.64, 0.89, 1.00, 0.88, 1.00, 1.00, 1.00, 0.98, 0.90, 1.00, 1.00, 0.55, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.07, 0.26, 0.05, 0.26, 0.26, 0.26, 0.23, 0.08, 0.93, 1.00, 0.42, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.33, 0.99, 0.82, 0.07, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.09, 0.91, 1.00, 0.33, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.51, 1.00, 0.93, 0.17, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.23, 0.98, 1.00, 0.24, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.52, 1.00, 0.73, 0.02, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.04, 0.80, 0.97, 0.23, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.49, 1.00, 0.71, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.29, 0.98, 0.94, 0.22, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.07, 0.87, 1.00, 0.65, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.01, 0.80, 1.00, 0.86, 0.14, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.15, 1.00, 1.00, 0.30, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.12, 0.88, 1.00, 0.45, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.52, 1.00, 1.00, 0.20, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.24, 0.95, 1.00, 1.00, 0.20, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.47, 1.00, 1.00, 0.86, 0.16, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.47, 1.00, 0.81, 0.07, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00,
|
||||
0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00
|
||||
};
|
||||
const int mnist_label = 7;
|
||||
@@ -0,0 +1,96 @@
|
||||
#include <transform.h>
|
||||
#include <stdio.h>
|
||||
|
||||
#include "tensorflow/lite/micro/all_ops_resolver.h"
|
||||
#include "tensorflow/lite/micro/micro_error_reporter.h"
|
||||
#include "tensorflow/lite/micro/micro_interpreter.h"
|
||||
#include "tensorflow/lite/schema/schema_generated.h"
|
||||
#include "tensorflow/lite/version.h"
|
||||
|
||||
#include "digit.h"
|
||||
#include "model.h"
|
||||
|
||||
namespace {
|
||||
tflite::ErrorReporter* error_reporter = nullptr;
|
||||
const tflite::Model* model = nullptr;
|
||||
tflite::MicroInterpreter* interpreter = nullptr;
|
||||
TfLiteTensor* input = nullptr;
|
||||
TfLiteTensor* output = nullptr;
|
||||
constexpr int kTensorArenaSize = 110 * 1024;
|
||||
uint8_t *tensor_arena = nullptr;
|
||||
//uint8_t tensor_arena[kTensorArenaSize];
|
||||
}
|
||||
|
||||
extern "C" void mnist_app() {
|
||||
tflite::MicroErrorReporter micro_error_reporter;
|
||||
error_reporter = µ_error_reporter;
|
||||
|
||||
model = tflite::GetModel(mnist_model);
|
||||
if (model->version() != TFLITE_SCHEMA_VERSION) {
|
||||
TF_LITE_REPORT_ERROR(error_reporter,
|
||||
"Model provided is schema version %d not equal "
|
||||
"to supported version %d.",
|
||||
model->version(), TFLITE_SCHEMA_VERSION);
|
||||
return;
|
||||
}
|
||||
|
||||
tensor_arena = (uint8_t *)malloc(kTensorArenaSize);
|
||||
if (tensor_arena == nullptr) {
|
||||
TF_LITE_REPORT_ERROR(error_reporter, "malloc for tensor_arena failed");
|
||||
return;
|
||||
}
|
||||
|
||||
tflite::AllOpsResolver resolver;
|
||||
tflite::MicroInterpreter static_interpreter(
|
||||
model, resolver, tensor_arena, kTensorArenaSize, error_reporter);
|
||||
interpreter = &static_interpreter;
|
||||
|
||||
// Allocate memory from the tensor_arena for the model's tensors.
|
||||
TfLiteStatus allocate_status = interpreter->AllocateTensors();
|
||||
if (allocate_status != kTfLiteOk) {
|
||||
TF_LITE_REPORT_ERROR(error_reporter, "AllocateTensors() failed");
|
||||
return;
|
||||
}
|
||||
|
||||
input = interpreter->input(0);
|
||||
output = interpreter->output(0);
|
||||
|
||||
printf("------- Input Digit -------\n");
|
||||
for (int i = 0; i < 28; i++) {
|
||||
for (int j = 0; j < 28; j++) {
|
||||
if (mnist_digit[i*28+j] > 0.3)
|
||||
printf("#");
|
||||
else
|
||||
printf(".");
|
||||
}
|
||||
printf("\n");
|
||||
}
|
||||
|
||||
for (int i = 0; i < 28*28; i++) {
|
||||
input->data.f[i] = mnist_digit[i];
|
||||
}
|
||||
|
||||
TfLiteStatus invoke_status = interpreter->Invoke();
|
||||
if (invoke_status != kTfLiteOk) {
|
||||
TF_LITE_REPORT_ERROR(error_reporter, "Invoke failed on x_val\n");
|
||||
return;
|
||||
}
|
||||
|
||||
// Read the predicted y value from the model's output tensor
|
||||
float max = 0.0;
|
||||
int index;
|
||||
for (int i = 0; i < 10; i++) {
|
||||
if(output->data.f[i]>max){
|
||||
max = output->data.f[i];
|
||||
index = i;
|
||||
}
|
||||
}
|
||||
printf("------- Output Result -------\n");
|
||||
printf("result is %d\n", index);
|
||||
}
|
||||
|
||||
extern "C" {
|
||||
#ifdef __RT_THREAD_H__
|
||||
MSH_CMD_EXPORT(mnist_app, run mnist app);
|
||||
#endif
|
||||
}
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,36 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
import tensorflow as tf
|
||||
|
||||
print("TensorFlow version %s" % (tf.__version__))
|
||||
|
||||
def show(image):
|
||||
for i in range(28):
|
||||
for j in range(28):
|
||||
if image[i][j] > 0.3:
|
||||
print('#', end = '')
|
||||
else:
|
||||
print('.', end = '')
|
||||
print()
|
||||
|
||||
digit_file_path = 'digit.h'
|
||||
digit_content = '''const float mnist_digit[] = {
|
||||
%s
|
||||
};
|
||||
const int mnist_label = %d;
|
||||
'''
|
||||
|
||||
if __name__ == '__main__':
|
||||
mnist = tf.keras.datasets.mnist
|
||||
(_, _), (test_images, test_labels) = mnist.load_data()
|
||||
index = 0
|
||||
shape = 28
|
||||
image = test_images[index].astype('float32')/255
|
||||
label = test_labels[index]
|
||||
print('label: %d' % label)
|
||||
#show(image)
|
||||
digit_data = (',\n ').join([ (', ').join([ '%.2f' % image[row][col] for col in range(shape)]) for row in range(shape)])
|
||||
digit_file = open(digit_file_path, 'w')
|
||||
digit_file.write(digit_content % (digit_data, label))
|
||||
digit_file.close()
|
||||
|
||||
@@ -0,0 +1,23 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
#tflite_file_path = 'mnist-default-quan.tflite'
|
||||
tflite_file_path = 'mnist.tflite'
|
||||
model_file_path = 'model.h'
|
||||
|
||||
tflite_file = open(tflite_file_path, 'rb')
|
||||
tflite_data = tflite_file.read()
|
||||
tflite_file.close()
|
||||
tflite_array = [ '0x%02x' % byte for byte in tflite_data ]
|
||||
|
||||
model_content = '''unsigned char mnist_model[] = {
|
||||
%s
|
||||
};
|
||||
unsigned int mnist_model_len = %d;
|
||||
'''
|
||||
# 12 bytes in a line, the same with xxd
|
||||
bytes_of_line = 12
|
||||
model_data = (',\n ').join([ (', ').join(tflite_array[i:i+bytes_of_line]) for i in range(0, len(tflite_array), bytes_of_line) ])
|
||||
model_file = open(model_file_path, 'w')
|
||||
model_file.write(model_content % (model_data, len(tflite_array)))
|
||||
model_file.close()
|
||||
|
||||
@@ -0,0 +1,40 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
import tensorflow as tf
|
||||
|
||||
print("TensorFlow version %s" % (tf.__version__))
|
||||
|
||||
MODEL_NAME_H5 = 'mnist.h5'
|
||||
MODEL_NAME_TFLITE = 'mnist.tflite'
|
||||
DEFAULT_QUAN_MODEL_NAME_TFLITE = 'mnist-default-quan.tflite'
|
||||
FULL_QUAN_MODEL_NAME_TFLITE = 'mnist-full-quan.tflite'
|
||||
|
||||
|
||||
def show(image):
|
||||
for i in range(28):
|
||||
for j in range(28):
|
||||
if image[i][j][0] > 0.3:
|
||||
print('#', end = '')
|
||||
else:
|
||||
print(' ', end = '')
|
||||
print()
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
mnist = tf.keras.datasets.mnist
|
||||
(_, _), (test_images, test_labels) = mnist.load_data()
|
||||
test_images = test_images.reshape(10000, 28, 28, 1)
|
||||
index = 0
|
||||
input_image = test_images[index].astype('float32')/255
|
||||
target_label = test_labels[index]
|
||||
|
||||
interpreter = tf.lite.Interpreter(model_path = DEFAULT_QUAN_MODEL_NAME_TFLITE)
|
||||
interpreter.allocate_tensors()
|
||||
input_details = interpreter.get_input_details()[0]
|
||||
output_details = interpreter.get_output_details()[0]
|
||||
interpreter.set_tensor(input_details['index'], [input_image])
|
||||
interpreter.invoke()
|
||||
output = interpreter.get_tensor(output_details['index'])[0]
|
||||
|
||||
show(input_image)
|
||||
print('target label: %d, predict label: %d' % (target_label, output.argmax()))
|
||||
@@ -0,0 +1,109 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
import os
|
||||
import tensorflow as tf
|
||||
|
||||
print("TensorFlow version %s" % (tf.__version__))
|
||||
|
||||
MODEL_NAME_H5 = 'mnist.h5'
|
||||
MODEL_NAME_TFLITE = 'mnist.tflite'
|
||||
DEFAULT_QUAN_MODEL_NAME_TFLITE = 'mnist-default-quan.tflite'
|
||||
FULL_QUAN_MODEL_NAME_TFLITE = 'mnist-full-quan.tflite'
|
||||
|
||||
def build_model(model_name):
|
||||
print('\n>>> load mnist dataset')
|
||||
mnist = tf.keras.datasets.mnist
|
||||
(train_images, train_labels),(test_images, test_labels) = mnist.load_data()
|
||||
print("train images shape: ", train_images.shape)
|
||||
print("train labels shape: ", train_labels.shape)
|
||||
print("test images shape: ", test_images.shape)
|
||||
print("test labels shape: ", test_labels.shape)
|
||||
|
||||
# transform label to categorical, like: 2 -> [0, 0, 1, 0, 0, 0, 0, 0, 0, 0]
|
||||
print('\n>>> transform label to categorical')
|
||||
train_labels = tf.keras.utils.to_categorical(train_labels)
|
||||
test_labels = tf.keras.utils.to_categorical(test_labels)
|
||||
print("train labels shape: ", train_labels.shape)
|
||||
print("test labels shape: ", test_labels.shape)
|
||||
|
||||
# transform color like: [0, 255] -> 0.xxx
|
||||
print('\n>>> transform image color into float32')
|
||||
train_images = train_images.astype('float32') / 255
|
||||
test_images = test_images.astype('float32') / 255
|
||||
|
||||
# reshape image like: (60000, 28, 28) -> (60000, 28, 28, 1)
|
||||
print('\n>>> reshape image with color channel')
|
||||
train_images = train_images.reshape((60000, 28, 28, 1))
|
||||
test_images = test_images.reshape((10000, 28, 28, 1))
|
||||
print("train images shape: ", train_images.shape)
|
||||
print("test images shape: ", test_images.shape)
|
||||
|
||||
print('\n>>> build model')
|
||||
model = tf.keras.models.Sequential([
|
||||
tf.keras.layers.Conv2D(32, (3, 3), activation=tf.nn.relu, input_shape=(28, 28, 1)),
|
||||
tf.keras.layers.MaxPooling2D((2, 2)),
|
||||
tf.keras.layers.Conv2D(64, (3, 3), activation=tf.nn.relu),
|
||||
tf.keras.layers.MaxPooling2D((2, 2)),
|
||||
tf.keras.layers.Conv2D(64, (3, 3), activation=tf.nn.relu),
|
||||
tf.keras.layers.Flatten(),
|
||||
tf.keras.layers.Dense(64, activation=tf.nn.relu),
|
||||
tf.keras.layers.Dense(10, activation=tf.nn.softmax)
|
||||
])
|
||||
model.compile(optimizer='rmsprop',
|
||||
loss='categorical_crossentropy',
|
||||
metrics=['accuracy'])
|
||||
model.summary()
|
||||
|
||||
print('\n>>> train the model')
|
||||
early_stopping = tf.keras.callbacks.EarlyStopping(
|
||||
monitor='loss', min_delta=0.0005, patience=3, verbose=1, mode='auto',
|
||||
baseline=None, restore_best_weights=True
|
||||
)
|
||||
model.fit(train_images, train_labels, epochs=100, batch_size=64, callbacks=[early_stopping])
|
||||
|
||||
print('\n>>> evaluate the model')
|
||||
test_loss, test_acc = model.evaluate(test_images, test_labels)
|
||||
print("lost: %f, accuracy: %f" % (test_loss, test_acc))
|
||||
|
||||
print('\n>>> save the keras model as %s' % model_name)
|
||||
model.save(model_name)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
|
||||
if not os.path.exists(MODEL_NAME_H5):
|
||||
build_model(MODEL_NAME_H5)
|
||||
|
||||
if not os.path.exists(MODEL_NAME_TFLITE):
|
||||
print('\n>>> save the tflite model as %s' % MODEL_NAME_TFLITE)
|
||||
converter = tf.lite.TFLiteConverter.from_keras_model(tf.keras.models.load_model(MODEL_NAME_H5))
|
||||
tflite_model = converter.convert()
|
||||
with open(MODEL_NAME_TFLITE, "wb") as f:
|
||||
f.write(tflite_model)
|
||||
|
||||
if not os.path.exists(DEFAULT_QUAN_MODEL_NAME_TFLITE):
|
||||
print('\n>>> save the default quantized model as %s' % DEFAULT_QUAN_MODEL_NAME_TFLITE)
|
||||
converter = tf.lite.TFLiteConverter.from_keras_model(tf.keras.models.load_model(MODEL_NAME_H5))
|
||||
converter.optimizations = [tf.lite.Optimize.DEFAULT]
|
||||
tflite_model = converter.convert()
|
||||
with open(DEFAULT_QUAN_MODEL_NAME_TFLITE, "wb") as f:
|
||||
f.write(tflite_model)
|
||||
|
||||
if not os.path.exists(FULL_QUAN_MODEL_NAME_TFLITE):
|
||||
mnist = tf.keras.datasets.mnist
|
||||
(train_images, _), (_, _) = mnist.load_data()
|
||||
train_images = train_images.astype('float32') / 255
|
||||
train_images = train_images.reshape((60000, 28, 28, 1))
|
||||
def representative_data_gen():
|
||||
for input_value in tf.data.Dataset.from_tensor_slices(train_images).batch(1).take(100):
|
||||
yield [input_value]
|
||||
print('\n>>> save the full quantized model as %s' % DEFAULT_QUAN_MODEL_NAME_TFLITE)
|
||||
converter = tf.lite.TFLiteConverter.from_keras_model(tf.keras.models.load_model(MODEL_NAME_H5))
|
||||
converter.optimizations = [tf.lite.Optimize.DEFAULT]
|
||||
converter.representative_dataset = representative_data_gen
|
||||
converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
|
||||
converter.inference_input_type = tf.uint8
|
||||
converter.inference_output_type = tf.uint8
|
||||
tflite_model = converter.convert()
|
||||
with open(FULL_QUAN_MODEL_NAME_TFLITE, "wb") as f:
|
||||
f.write(tflite_model)
|
||||
Reference in New Issue
Block a user