同一个模型，python下模型初始化成果，c++下失败 - Toybrick

#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>
#include <fstream>
#include <iostream>
#include <algorithm>
#include <queue>
#include <sys/time.h>
#include "rknn_api.h"
#include "opencv2/core/core.hpp"
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/highgui/highgui.hpp"
using namespace std;
int main(int argc, char** argv)
{
const char *img_path = "../tmp/stand-06_13_02-01374.jpg";
const char *model_path = "../tmp/rensnet50.rknn";
const int output_elems = 2;
const int img_width = 224;
const int img_height = 224;
const int img_channels = 3;
const int input_index = 0; // node name "input" 输入输出节点的索引？？
const int output_index = 0; // node name "MobilenetV1/Predictions/Reshape_1"
// Load image
cv::Mat img = cv::imread(img_path, 1);
if(!img.data) {
printf("cv::imread %s fail!\n", img_path);
return -1;
}
if(img.cols != img_width || img.rows != img_height)
cv::resize(img, img, cv::Size(img_width, img_height), (0, 0), (0, 0), cv::INTER_LINEAR);
//BGR->RGB
//cv::cvtColor(img, img, cv::COLOR_BGR2RGB);
// Load model
FILE *fp = fopen(model_path, "rb");
if(fp == NULL) {
printf("fopen %s fail!\n", model_path);
return -1;
}
fseek(fp, 0, SEEK_END); //重定位流(数据流/文件)上的文件内部位置指针
int model_len = ftell(fp); //函数 ftell() 用于得到文件位置指针当前位置相对于文件首的偏移字节数
void *model = malloc(model_len);
fseek(fp, 0, SEEK_SET); //SEEK_SET开头－0，SEEK_CUR当前位置－1，SEEK_END结束－2
if(model_len != fread(model, 1, model_len, fp)) { //权重读到model中
printf("fread %s fail!\n", model_path);
free(model);
return -1;
}
// Start Inference
rknn_input inputs[1]; //输入结构体，index,buf:指向数据的指针,size,pass_through:数据是否需要转换,type:数据的类型,fmt:数据的格式
rknn_output outputs[1]; //want_float,is_prealloc,index,buf,size
rknn_tensor_attr output0_attr; //index,n_dims,dims,name,n_elems,size,fmt:NCHW-NHWC,type,qnt_type,f1,zp,scale
int ret = 0;
rknn_context ctx = 0; //模型的对象？
ret = rknn_init(&ctx, model, model_len, RKNN_FLAG_PRIOR_MEDIUM | RKNN_FLAG_COLLECT_PERF_MASK); //中优先级，打开测试开关
if(ret < 0) {
printf("rknn_init fail! ret=%d\n", ret);
goto Error;
}
output0_attr.index = 0; //设置
ret = rknn_query(ctx, RKNN_QUERY_OUTPUT_ATTR, &output0_attr, sizeof(output0_attr)); //查询ctx对象输出属性
if(ret < 0) {
printf("rknn_query fail! ret=%d\n", ret);
goto Error;
}
inputs[0].index = input_index; //这个索引搞得不是很清楚
inputs[0].buf = img.data; //装载数据
inputs[0].size = img_width * img_height * img_channels; //占内存大小
inputs[0].pass_through = false;
inputs[0].type = RKNN_TENSOR_UINT8;
inputs[0].fmt = RKNN_TENSOR_NHWC;
ret = rknn_inputs_set(ctx, 1, inputs); //对象，输入的个数，输入的数组
if(ret < 0) {
printf("rknn_input_set fail! ret=%d\n", ret);
goto Error;
}
ret = rknn_run(ctx, nullptr);
if(ret < 0) {
printf("rknn_run fail! ret=%d\n", ret);
goto Error;
}
outputs[0].want_float = true;
outputs[0].is_prealloc = false;
ret = rknn_outputs_get(ctx, 1, outputs, nullptr); //从ctx中将output取出
if(ret < 0) {
printf("rknn_outputs_get fail! ret=%d\n", ret);
goto Error;
}
// Process output
if(outputs[0].size == output0_attr.n_elems * sizeof(float))
{
for(int i = 0;i <output0_attr.n_elems;i++){
std::cout << "outputs " << i << " : " << ((float*)outputs[0].buf)[i] << std::endl;
}
std::cout << "outputs[0].size : " << outputs[0].size << std::endl;
std::cout << "outputs0_attr.n_elems : " << output0_attr.n_elems << std::endl;
}
else
{
printf("rknn_outputs_get fail! get output_size = [%d], but expect %u!\n",
outputs[0].size, (uint32_t)(output0_attr.n_elems * sizeof(float)));
}
rknn_outputs_release(ctx, 1, outputs);
Error:
if(ctx > 0) rknn_destroy(ctx);
if(model) free(model);
if(fp) fclose(fp);
return 0;
}