osmr / facedetver

Face Detection Verifier

Face Detection Verifier

This repo contains scripts for solving the binary classification problem, where the positive class is undistorted images of a person’s face, and the negative class is everything else, including images of parts of a person’s face, face drawings, etc.

The repository contains scripts for building and analyzing the corresponding dataset, for training and testing models. Two deep learning frameworks are supported: MXNet/Gluon and PyTorch. All scripts are completely duplicated. In addition, the releases contain two training datasets and four models that solve the problem.

Deployment Instructions

Recommended repository deployment protocol on the machine with CUDA 10.0 and cuDNN 7:

1. Install prerequesities for MXNet:

   apt update
   apt upgrade
   apt install -y htop mc wget unzip python3-pip ipython3
   apt install -y build-essential git ninja-build ccache
   apt install -y apt-transport-https build-essential ca-certificates curl git libatlas-base-dev libcurl4-openssl-dev libjemalloc-dev libhdf5-dev liblapack-dev libopenblas-dev libopencv-dev libturbojpeg libzmq3-dev ninja-build software-properties-common sudo vim-nox
   apt install -y libopenblas-dev libopencv-dev
   apt install -y libsm6 libxext6 libxrender-dev
   

2. Install pip-packages from requirements.txt:

   pip3 install --upgrade numpy opencv-python imgaug tqdm
   pip3 install --upgrade mxnet-cu100 gluoncv2
   pip3 install --upgrade torch torchvision pytorchcv
   

3. Clone the repo:

   mkdir projects
   cd projects
   git clone https://github.com/osmr/facedetver.git
   

4. Create directory for dataset and models:

   mkdir facedetver_data
   cd facedetver_data
   

5. Download and extract dataset FDV1:

   mkdir fdv1
   cd fdv1
   wget https://github.com/osmr/facedetver/releases/download/v0.0.1/fdv1_test.zip
   wget https://github.com/osmr/facedetver/releases/download/v0.0.1/fdv1_train.zip
   wget https://github.com/osmr/facedetver/releases/download/v0.0.1/fdv1_val.zip
   unzip fdv1_test.zip
   unzip fdv1_train.zip
   unzip fdv1_val.zip
   

6. Or download and extract dataset FDV2:

   mkdir fdv2
   cd fdv2
   wget https://github.com/osmr/facedetver/releases/download/v0.0.2/fdv2_test.zip
   wget https://github.com/osmr/facedetver/releases/download/v0.0.2/fdv2_train.zip
   wget https://github.com/osmr/facedetver/releases/download/v0.0.2/fdv2_val.zip
   unzip fdv2_test.zip
   unzip fdv2_train.zip
   unzip fdv2_val.zip
   

7. Download and extract a model:

   cd ..
   mkdir resnet18_fdv1-0014
   cd resnet18_fdv1-0014
   wget https://github.com/osmr/facedetver/releases/download/v0.0.3/resnet18_fdv1-0014-a03f116e.params.zip
   unzip resnet18_fdv1-0014-a03f116e.params.zip
   

8. Run a testing script:

   cd ../../facedetver
   python3 eval_gl.py --num-gpus=1 --model=resnet18 --save-dir=../facedetver_data/resnet18_fdv1-0014/ --batch-size=100 -j=4 --resume=../facedetver_data/resnet18_fdv1-0014/resnet18_fdv1-0014-a03f116e.params --calc-flops --show-bad-samples --data-subset=test
   

Pretrainded Models

Model	Dataset	Framework	Acc	F1	MCC	Params	FLOPs/2	Remarks
ResNet-18	FDV1	Gluon	0.9976	0.9976	0.9952	11,177,538	1,819.90M	Training (log)
ResNet-18	FDV1	PyTorch	0.9976	0.9976	0.9952	11,177,538	1,819.90M	Training (log)
ResNet-18	FDV2	Gluon	0.9971	0.9971	0.9942	11,177,538	1,819.90M	Training (log)
ResNet-18	FDV2	PyTorch	0.9971	0.9971	0.9942	11,177,538	1,819.90M	Training (log)