Face Detection Based on AdaBoost Algorithm

机器学习 实验

你可以点击这里查看中文版本。

Motivation of Experiment

1. Understand Adaboost further
2. Get familiar with the basic method of face detection
3. Learn to use Adaboost to solve the face detection problem,and combine the theory with the actual project
4. Experience the complete process of machine learning

Dataset

1. This experiment provides 1000 pictures, of which 500 are human face RGB images, stored in datasets/original/face; the other 500 are non-face RGB images, stored in datasets/original/nonface.
2. The dataset is included in the example repository. Download it and divide it into training set and validation set.

Environment for Experiment

• python3, at least including following python package: sklearn, numpy, pickle, PIL, opencv-python.
  It is recommended to install anaconda3 directly, which has built-in python package above except opencv-python.You can use pip to install opencv-python from Tsinghua Open Source Mirror:

pip install -i https://pypi.tuna.tsinghua.edu.cn/simple opencv-python

• PyCharm Community Integrated Development Environment (optional).

Time and Place

2018-10-14 2:00-5:00 PM B7-138(Mingkui Tan) B7-238（Qingyao Wu）

Submit Deadline

2018-11-18 12:00 AM

Experiment Form

Complete Independently.

Experiment Steps

Face Classification
1. Load data set data. The images are supposed to converted into grayscale images with size of 24 * 24, the number and the proportion of the positive and negative samples is not limited, the data set label is not limited.
2. Processing data set data to extract NPD features. Extract features using the NPDFeature class in feature.py. (Tip: Because the time of the pretreatment is relatively long, it can be pretreated with pickle function library dump() save the data in the cache, then may be used load() function reads the characteristic data from cache.)
3. The data set is divisded into training set and calidation set, this experiment does not divide the test set.
4. Write all AdaBoostClassifier functions based on the reserved interface in ensemble.py. The following is the guide of fit function in the AdaBoostClassifier class:
    4.1 Initialize training set weights $\omega$, each training sample is given the same weight.
    4.2 Training a base classifier , which can be sklearn.tree library DecisionTreeClassifier (note that the training time you need to pass the weight $\omega$ as a parameter).
    4.3 Calculate the classification error rate $\epsilon$ of the base classifier on the training set.
    4.4 Calculate the parameter $\alpha$ according to the classification error rate $\epsilon$.
    4.5 Update training set weights $\omega$.
    4.6 Repeat steps 4.2-4.6 above for iteration, the number of iterations is based on the number of classifiers.
5. Predict and verify the accuracy on the validation set using the method in AdaBoostClassifier and use classification_report () of the sklearn.metrics library function writes predicted result to classifier_report.txt .

Face Detection

1. Run the face_detection.py file. Experience the OpenCV's built-in method of face detection using Haar Feature-based Cascade Classifiers.The result will be save as detect_result.jpg.
2. You can provide your own images to replace the default test image.

Finishing experiment report according to results(The template of report can be found in the example repository.

Evaluation

Requirement for Submission

Submission process

1. Access 222.201.187.50:7001.
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.

References

[1] Liao, S., Jain, A. K., & Li, S. Z. (2016). A fast and accurate unconstrained face detector. IEEE transactions on pattern analysis and machine intelligence, 38(2), 211-223.
[2] 周志华. 机器学习. 北京：清华大学出版社，2016：173-177