建立虛擬環境
CPU 環境
conda create --name myenv36_cpu -y python=3.6 && \
source activate myenv36_cpu && \
conda install --yes --file requirements.txt
GPU 環境
conda create --name myenv36_gpu -y python=3.6 && \
source activate myenv36_gpu && \
conda install --yes --file requirements.txt
若套件發生問題 手動安裝方式
cv2
No module named 'cv2'
手動安裝
conda install opencv
easydict
No module named 'easydict'
手動安裝
pip uninstall easydict
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple --default-timeout=1000 EasyDict
torch
No module named 'torch'
手動安裝
conda install pytorch==0.4.0 torchvision==0.2.1 -c pytorch
gpu version CUDA 10.0
conda install pytorch==1.2.0 torchvision==0.4.0 cudatoolkit=10.0 -c pytorch
conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch
bcolz
No module named 'bcolz'
手動安裝
conda install bcolz==1.2.1
mxnet
No module named 'mxnet'
手動安裝
conda install mxnet
tqdm
No module named 'tqdm'
手動安裝
conda install tqdm==4.23.4
sklearn
No module named 'sklearn'
手動安裝
conda install scikit-learn
tensorboardx
No module named 'tensorboardx'
手動安裝
conda install -c conda-forge/label/gcc7 tensorboardx
matplotlib
No module named 'matplotlib'
手動安裝
conda install matplotlib
其他錯誤
cv2.imshow("My Capture",frame_text) 執行錯誤
pip install opencv-contrib-python
下載資料集
[dropbox](https://www.dropbox.com/s/wpx6tqjf0y5mf6r/faces_ms1m-refine-v2_112x112.zip?dl=0
解壓縮放置 data 資料夾
執行資料解壓縮
python prepare_data.py
將數據及資料解壓轉存到 data/faces_emore/images 底下
相機取的使用者
python take_pic.py -n David
按下t拍照
識別測試
python face_verify.py
這邊需要訓練模型
Training
python train.py -net mobilefacenet -b 200 -w 4
測試機器記憶體只有 15.6g，修改 batch_size
python train.py -net mobilefacenet -b 96 -w 4
參考資料
How to install the NVIDIA drivers on Ubuntu 20.04 Focal Fossa Linux PyTorch