65 lines
2.9 KiB
Markdown
65 lines
2.9 KiB
Markdown
origin github rep: https://github.com/serengil/deepface
|
||
(推荐：直接使用我提交的仓库代码，并按照教程安装。如果需要查看更改原作者的仓库，请注意：
|
||
我拉取的作者仓库的`commit`版本是： commit SHA: 09f325bc3706b8be70ab509a5a00843a6b6c5cdb ，请拉取或查看相同`commit`版本的原作者代码)
|
||
|
||
|
||
# model store
|
||
model checkpoint 放在了 `C:\Users\User\.deepface\weights`
|
||
|
||
# installation and setup (安装并启动)
|
||
```
|
||
$ conda create -n [your-env-name] python=3.7.16
|
||
$ conda activate [your-env-name]
|
||
$ cd ./deepface
|
||
$ pip install -e .
|
||
$ pip install "uvicorn[standard]"
|
||
|
||
$ cd ..
|
||
$ uvicorn webmain:app --port [your-port] --host [your-host]
|
||
```
|
||
|
||
web服务的url路径
|
||
```
|
||
http://your-host:your-port/docs#/
|
||
```
|
||
|
||
# Model introduce
|
||
```
|
||
人脸识别所用的模型
|
||
推荐使用(推荐优先级同顺序)：GhostFaceNet, ArcFace
|
||
Deepface is a hybrid face recognition package. It currently wraps many state-of-the-art face recognition models:
|
||
model_name Declared LFW Score 是否可用 备注
|
||
VGG-Face 98.9% 可用
|
||
Facenet 99.2% 可用
|
||
Facenet512 99.6% 可用 效果似乎不好
|
||
OpenFace 92.9% 可用
|
||
DeepID 97.4% 可用
|
||
Dlib 99.3% 不可用
|
||
SFace 99.5% 可用
|
||
ArcFace 99.5% 可用 重点测试**（阈值应该可以设置小一点）
|
||
GhostFaceNet 99.7% 可用 重点测试**
|
||
Human-beings 97.5% 不可用
|
||
|
||
The default configuration uses GhostFaceNet model.
|
||
```
|
||
|
||
```
|
||
推荐(优先级同顺序)：retinaface, mtcnn, opencv, ssd
|
||
backends = {
|
||
"opencv": OpenCv.OpenCvClient, 可用，效果一般，容易找不到人脸
|
||
"mtcnn": MtCnn.MtCnnClient, 可用，效果挺好
|
||
"ssd": Ssd.SsdClient, 可用，容易找到太多人脸
|
||
"dlib": Dlib.DlibClient, 不可用，Dlib is an optional detector, ensure the library is installed.Please install using 'pip install dlib'
|
||
"retinaface": RetinaFace.RetinaFaceClient, 可用，效果挺好
|
||
"mediapipe": MediaPipe.MediaPipeClient, 未知，MediaPipe is an optional detector, ensure the library is installed.Please install using 'pip install mediapipe'
|
||
"yolov8": Yolo.YoloClient, 不可用，Yolo is an optional detector, ensure the library is installed. Please install using 'pip install ultralytics'
|
||
"yunet": YuNet.YuNetClient, 未知，没下载model ck
|
||
"fastmtcnn": FastMtCnn.FastMtCnnClient, 不可用，FastMtcnn is an optional detector, ensure the library is installed.Please install using 'pip install facenet-pytorch'
|
||
}
|
||
```
|
||
|
||
```
|
||
这是计算不同人脸相似度时使用的函数，直接使用代码中默认的既可，无需更改
|
||
distance_metric = ['euclidean_l2','cosine','euclidean']
|
||
author: Euclidean L2 form seems to be more stable than cosine and regular Euclidean distance based on experiments.
|
||
``` |