【人脸检测】复现官方RetinaFace

刺骨的言语ヽ痛彻心扉 2022-11-21 09:51 263阅读 0赞

[https://github.com/deepinsight/insightface/tree/master/detection/RetinaFace][https_github.com_deepinsight_insightface_tree_master_detection_RetinaFace]

## 0.环境 ##

mxnet-cu90 @ file:mxnet_cu90-1.2.0-py2.py3-none-manylinux1_x86_64.whl
    Cython==0.29.21
    easydict==1.9
    numpy==1.14.6
    opencv-python==4.4.0.44

环境配置步骤：

git clone https://github.com/deepinsight/insightface.git
    cd insightface/RetinaFace
    
    pip install -i https://pypi.tuna.tsinghua.edu.cn/simple mxnet_cu90-1.2.0-py2.py3-none-manylinux1_x86_64.whl
    pip install -i https://pypi.tuna.tsinghua.edu.cn/simple Cython==0.29.21 \
    easydict==1.9 \
    numpy==1.14.6 \
    opencv-python==4.4.0.44
    
    make

## 1.模型与数据准备 ##

（1）模型准备

ImageNet ResNet50 ([baidu cloud][] and [dropbox][]).

ImageNet ResNet152 ([baidu cloud][baidu cloud 1] and [dropbox][dropbox 1]).

将模型解压放到model文件下。

（2）数据准备

原始数据：train+val+test [http://shuoyang1213.me/WIDERFACE/WiderFace\_Results.html][http_shuoyang1213.me_WIDERFACE_WiderFace_Results.html]

标签数据：[https://pan.baidu.com/s/1Laby0EctfuJGgGMgRRgykA][https_pan.baidu.com_s_1Laby0EctfuJGgGMgRRgykA]

按照官方给的格式放：

data/retinaface/
        train/
          images/
          label.txt
        val/
          images/
          label.txt
        test/
          images/
          label.txt

## 2.训练--Resnet50 ##

如果按照0中的将环境配置好后，直接运行下面就可以。

CUDA_VISIBLE_DEVICES='0,1,2,3' python -u train.py --prefix ./model/retina --network resnet

运行后，输出配置信息：

![20201031142745484.png][]

[https_github.com_deepinsight_insightface_tree_master_detection_RetinaFace]: https://github.com/deepinsight/insightface/tree/master/detection/RetinaFace
[baidu cloud]: https://pan.baidu.com/s/1WAkU9ZA_j-OmzO-sdk9whA
[dropbox]: https://www.dropbox.com/s/48b850vmnaaasfl/imagenet-resnet-50.zip?dl=0
[baidu cloud 1]: https://pan.baidu.com/s/1nzQ6CzmdKFzg8bM8ChZFQg
[dropbox 1]: https://www.dropbox.com/s/8ypcra4nqvm32v6/imagenet-resnet-152.zip?dl=0
[http_shuoyang1213.me_WIDERFACE_WiderFace_Results.html]: http://shuoyang1213.me/WIDERFACE/WiderFace_Results.html
[https_pan.baidu.com_s_1Laby0EctfuJGgGMgRRgykA]: https://pan.baidu.com/s/1Laby0EctfuJGgGMgRRgykA
[20201031142745484.png]: /images/20221120/3e07f2881fce49fc984abfe3f5acaa00.png