README

This repo serves as the code implementation of the project assigned in the module Applied Machine Learning Systems ELEC0134.

Description

In this project, we are required to tackle two pairs of tasks within the field of computer vision. The first pair of tasks deals with binary classification problem, covering gender detection (Task A1) and emotion detection (Task A2). The next one works on multiclass classification problem, covering face shape recognition (Task B1) and eye color recognition (Task B2) respectively. In this paper, traditional machine learning techniques like support vector machine (SVM) with linear kernel and random forest are involved to solve these image classification problems, along with two face detection approaches known as dlib’s facial landmark detector and Haar feature-based cascade classifier. The usage of these methods combined with a complete feature engineering process gives us good classification performance, where the accuracy score on testing datasets for Task A1 reaches 92.1% and Task A2 as 89.6%. In addition, Convolutional Neural Network (CNN), one of the state-of-the-art deep learning algorithms, has also shown its strong capability on these tasks which achieves the accuracy score of 94.5% to Task B1 and 80.3% to Task B2.

Results

Task	Model	Train Acc	Val Acc	Test Acc	Elapsed Time
A1	LinearSVM	93.5%	92.1%	92.1%	00:00:24
A2	LinearSVM	89.8%	88.9%	89.6%	00:00:16
B1	CNN	98.5%	94.4%	94.5%	00:19:08
B2	CNN	79.2%	81.1%	80.3%	00:15:40

Requirements

• Python 3.7+ (Python 3.7.9 is recommended which is the version used in the development.)
• macOS or Windows (For Windows users, please change '/' into '\\' in helper.py line 161.)
• Required modules
  • check requirement.txt to ensure all modules included have been installed
  • or run pip3 install -r requirements.txt in terminal to install with ease

Usage

1. Switch to the root directory of the repo
2. Run python3 main.py in terminal to start the main program
3. Monitor the running process in base_log.log if you like (logging level is set to INFO by default, feel free to alter)

Structure

├── A1                      # the folder of Task A1, where model and temporary dataset for Task A1 could be found
│   ├── A1.py               # scripts for Task A1
├── A2                      # the folder of Task A2, where model and temporary dataset for Task A2 could be found
│   ├── A2.py               # scripts for Task A2
├── B1                      # the folder of Task B1, where model for Task B1 could be found
│   ├── B1.py               # scripts for Task B1
├── B2                      # the folder of Task B2, where model for Task B2 could be found
│   ├── B2.py               # scripts for Task B2
├── Datasets                # the folder of datasets, with content removed on purpose
│   ├── Remark.md
│   ├── cartoon_set
│   ├── cartoon_set_test
│   ├── celeba
│   └── celeba_test
├── README.md               # this file
├── devlog                  # the folder of some experiment related files, kept for debugging
│   ├── candidate_models    # (for debugging only) trained models
│   ├── csv_datasets        # (for debugging only) temporary datasets
│   ├── devlog.md           # (for debugging only) records of the experiment results
│   ├── logs                # (for debugging only) logs
│   └── pics                # (for debugging only) pictures used for the report
├── helper                  # the folder of some necessary files
│   ├── base_log.log        # the main log, generated once the main program srarts
│   ├── face_models         # the models used for face detection
│   └── helper.py           # core functions for face detection, feature engineering and model training
├── main.py                 # the entrance to main program
└── requirements.txt        # the list of required modules

Having problems?

If you run into problems, please either file a github issue or send an email to [email protected].