基本思想：将adaface的mtcnn模型npy文件转成atlas310p模型进行推理。同时比对结果
ubuntu@ubuntu:~$ git clone https://github.com/mk-minchul/AdaFace.git
Cloning into 'AdaFace'...
remote: Enumerating objects: 236, done.
remote: Counting objects: 100% (109/109), done.
remote: Compressing objects: 100% (52/52), done.
remote: Total 236 (delta 67), reused 57 (delta 57), pack-reused 127 (from 1)
Receiving objects: 100% (236/236), 38.95 MiB | 9.06 MiB/s, done.
Resolving deltas: 100% (89/89), done.一、按照readme进行测试mtcnn模型，将main.py代码添加到/home/ubuntu/AdaFace/face_alignment/mtcnn_pytorch/目录下
from src import detect_faces
from PIL import Imageimage = Image.open('/home/ubuntu/AdaFace/face_alignment/test_images/img1.jpeg')
bounding_boxes, landmarks 
