Face Detection Based on Neural Network

机器学习 实验

点击这里获取中文版本

Motivation

1. Understand the basic theory of face detection using neural network.
2. Understand the processes of MTCNN and use it in practice.

Dataset for Training

When training PNet and Rnet, we use WiderFace, for face classification and face bounding box regression. When training ONet, we use WiderFace for face classification and face bounding box regression, and use Training Dataset for face feature point regression.

Environment for Experiment

anaconda3
pytorch 0.4.1
torchvision
opencv-python
tensorflow(only for python 3.4,3.5,3.6)

Time and Place

2018-11-24 8:50-12:15 AM B7-138(Mingkui Tan) B7-238（Qingyao Wu）

Submit Deadline

2018-12-29 12:00 noon

Experimental Form

Complete in group.

Experiment Steps

1.read the paper about MTCNN
2.run the given codes
1) get the codes
The complete codes for this experiment is given in MTCNN_pytorch, you can download it directly or using git clone command.

git clone https://github.com/wujiaju/mtcnn_pytorch.git

2) install the environment
1. download anaconda3 for your computer
2. use pip or conda to install pytorch 0.4.1 and torchvision
3. use pip or conda to install opencv-python

pip install http://download.pytorch.org/whl/cpu/torch-0.4.1-cp36-cp36m-win_amd64.whl     # Windows
pip install http://download.pytorch.org/whl/cpu/torch-0.4.1-cp36-cp36m-linux_x86_64.whl  # Linux
pip install torchvision
pip install opencv-python

3) test the given models
Use the model we have trained before to run on the test dataset, which is under mtcnn_pytorch/data/test_images/ and contains 64 pictures, input the command below and then go to mtcnn_pytorch/data/test_images/ to examine the result.

cd mtcnn_pytorch/
python test_image.py

4) training
NOTE： You should check whether the path of the training dataset is consistent with the path in the code. If not, you must modify the path in the code before you start training.

train PNet

cd mtcnn_pytorch
python preprocessing/gen_pnet_data.py
python preprocessing/assemble_pnet_imglist.py
python training/pnet/train.py

train RNet

cd mtcnn_pytorch
python preprocessing/gen_rnet_data.py
python preprocessing/assemble_rnet_imglist.py
python training/rnet/train.py

train ONet

cd mtcnn_pytorch
python preprocessing/gen_landmark_48.py
python preprocessing/gen_onet_data.py
python preprocessing/assemble_onet_imglist.py
python training/onet/train.py

After finishing training, you can examine the three trained models in mtcnn_pytorch/results/.

5) test your model

Use the given test dataset, which is under mtcnn_pytorch/data/test_images/ and contains 64 pictures, to test your model. Run the commands below and then examine the results in mtcnn_pytorch/data/you_result/.

cd mtcnn_pytorch/
python test_youModel_images.py

NOTE： You should check whether the path of the training dataset is consistent with the path in the code. If not, you must modify the path in the code before you start training.

Finishing experiment report according to result: The template of report can be found in example repository.

Experiment Report

The experiment report should contain your comprehension about MTCNN and the result of runing the given codes.

Evaluation

Requirement for Submission

Submission process

1.Access222.201.187.50:7001.
2.Click on the corresponding submission entry.
3.Fill in your name, student number, upload pdf format report and zip format code compression package.

Precautions

• Experiment reports and code can be uploaded multiple times, and multiple uploads will overwrite previously submitted files.
• After uploading, you can refresh the page and check if the upload is successful in the file list below.
• Teaching assistants save all uploaded results at the experimental deadline, and the files uploaded after the deadline are invalid.
• If you write an experiment report in Word, you need to export it to pdf format.
• The package format of the code file must be zip. Please do not submit the compressed file in rar format.
• Submit URL can only be accessed by campus network.
• The code is written in python language, the experimental report score standard English is better than Chinese, latex is better than word.

Any advice or idea is welcome to discuss with teaching assistant in QQ group.