zengwb-lx

FaceMOT

5

I trained a face tracking model with HRnet18, and the tracking performance was good (it could be tracked back within 1S after being lost）But not fast enough in the 2070S（about...

add_mask_one

8

你好，add_mask如果一张图有两个脸的话，无法同时给两个脸加上口罩，在遍历face_lms时会把前一次加的口罩去除掉，有什么好的建议修改加载一图多脸的情况吗？ 谢谢

Small data set

2

I'm going to train a object detection model， But my training set only has 50 pictures，Can you give me some good advice，

Regression loss coefficient.？？

1

why Regression loss coefficient. is 0.001, deep-head-pose set alpha is 1 or 2

input_map

1

你好， 我使用我的pytorch转的onnx模型，获取input_map时只获取了数据输入的节点input {'input0': name: "input0" type { tensor_type { elem_type: 1 shape { dim { dim_value: 1 } dim { dim_value: 3 } dim { dim_value: 112 } dim {...

Issues about mobileface

1

您好， 看到mobilefacenet-res2/4-***的精度都比mobilefacenet高很多，请问这些模型相对于mobilefacenet修改了什么？ 能开源训练代码给我吗？谢谢

about threshold

1

In the real world, How to set the optimal threshold for recognition？any tricks?

Thanks a lot for sharing amazing code!! But the Python version is still not fast enough, can you provide C++ version

使用MNN_FORWARD_VULKAN输出结果不对

3

1.使用的2.8.0版本预编译库，在window10 和android（adb方式）测试，代码如下：  在window10 和android（adb方式）使用cpu、opencl、cuda后端识别都是对的，使用vulkan结果却不对 人像风格迁移模型，麻烦帮忙看下什么问题 [cartoon_mnn.zip](https://github.com/alibaba/MNN/files/13768409/cartoon_mnn.zip)