Developed a comprehensive intelligent campus assistance system that combined calendar, word memorization, class schedule and other functions, using .NET and PYQT technology stack
Focus on learning functions(YOLOV8)：Users have to keep focuing on study. An alert will sound if the user leaves the computer
Model training Framework: Pytorch
Graphics card used: RTX3090
Dataset: Public data set on Feijian
Algorithm: ResNet-18 pre-trained on ImageNet (fully connected layer output changed to 4)
Optimizer: Adam (use Adam’s default learning rate in Pytorch)
Loss function: NLLLoss (compared to Cross Entropy Loss, it focuses more on “difficult” samples)
Number of training rounds: 30
Training curve(split the training set and test set into 4:1): Accuracy when epoch=30: 93.4%
Use early stopping and select the model with epoch=10
