hkdddld

Results 4 issues of hkdddld

我是在int MTCNNDetection()函数 while(cap.isOpened())后面用double start = ncnn::get_current_time();开始计时在该循环的末尾 double end = ncnn::get_current_time(); double time = end - start; double fps = 1000.0 / time; printf("Detect Time:%7.2f \n",fps);来计算该项目的fps通过给给个模型推理过程添加多线程目前最大只能达到4-5左右有人知道是为啥吗

circle(image_copy3, (contours2[i][0], contours2[i][1]), 2, Scalar(0, 255, 0), 2)是不是要改成circle(image_copy3, Point(contours2[i][j].x, contours2[i][j].y), 1, Scalar(0, 255, 0), 1);才能达到预期效果

![image](https://github.com/Tencent/ncnn/assets/154032974/58c0da78-168b-434f-b51c-2ad28c5a54fd) ./ncnn2table 在对下面det18执行时没有问题并成功生成了.table文件 ![image](https://github.com/Tencent/ncnn/assets/154032974/a7827759-f528-4155-be2a-f71468524553) 请问出现这种情况我需要怎么解决 images图片我是用的https://github.com/Tencent/ncnn/wiki/quantized-int8-inference提供的图片

### Describe the issue 我在推理fp16的yolov5模型时通过 ![微信图片_20240420200338](https://github.com/microsoft/onnxruntime/assets/154032974/6251030f-4f5a-4303-b22d-fac4f6bb287e) 这样转换出来推理不出结果是为什么 ### To reproduce ![微信图片_20240420200338](https://github.com/microsoft/onnxruntime/assets/154032974/2040cb44-c37f-4b07-9ed6-dbf94f721e25) ### Urgency _No response_ ### Platform Linux ### OS Version 22.0 ### ONNX Runtime Installation Built from Source ###...

quantization
stale