Mahjong Connect and Facial Expression Recognition

A course project for Artificial Intelligence, Autumn 2021 @ Department of Automation, Tsinghua

Lecturer: Rui Jiang

Mahjong Connect

A Automated Solver for 'Mahjong Connect', with GUI.

Download llkui.exe and put into 'MahjongConnect' folder to execute.

Facial Expression Recognition

File List:

Dir	FileName	Description (PT refers to Pre-training)
code	LeNet.ipynb	Original data, LeNet (Question 1)
	vgg9.ipynb	Original data, VGG-9 (Question 1)
	ResNet-11.ipynb	Original data, ResNet-11 (Question 1)
	vgg9-DataEnhance.ipynb	Enhanced data, VGG-9 (Question 2)
	vgg11aug_balance_pretrain.ipynb	Enhanced data, PT VGG-11
	ResNet18aug_balance_pretrain.ipynb	Enhanced data, PT ResNet-18
	ResNet18aug_balance_pretrain2.ipynb	Enhanced data, PT ResNet-18
	TestTime.ipynb	Test the running speed of each model's prediction mode
	myui.ui	QT file for the graphical interface
	myui.py	Runnable final program (Question 3)
model	vgg9net.pkl	Original data, VGG-9 model, Acc58.4%
	vgg9net_DataEnhance.pkl	Enhanced data, VGG-9 model, Acc58.8%
	finetune_vgg11net.pkl	Enhanced data, PT VGG-11, Acc62.1%
	finetune_resnet18.pkl	Enhanced data, PT ResNet-18, Acc59.1%
data	data.csv ; *.pt	Original data; Processed tensor files
media	N/A	Images/videos for testing the GUI program

Usage:

• First, perform model training and testing, there are two methods:

1. Download the entire data folder, run the .ipynb code in the code folder, at this time directly read data from the data folder;

2. Change the readtensor parameter in the .ipynb code to False, only put data.csv in the data folder, no need to download .pt files, at this time re-process data from data.csv (computationally time-consuming);

• Then perform data prediction testing, use the GUI program (Question 3):

1. Install the following packages in the python environment: pyqt5, opencv-python, torch, torchvision

2. Put the trained vgg9net.pkl in the model folder, then run the myui.py in the code folder

Download Data