Code for AI CUP 2020 Mango Image Recognition Challenge: Grade Classification and Defective Classification

pytorch torchvision cudatoolkit=10.1 rfconv 

1. Put your data (C1-P1_Train Dev_fixed) in this assignment folder 406410035_hw2_v1/
2. Change the path in options.py: ROOTPATH=[your absolute path to this assignment folder]
3. install all the requirements if you haven't yet by: while read requirement; do conda install --yes $requirement; done < requirement.txt
4. execute the program: python train.py if you have any other specification you can see the options in options.py to set in the command
5. after executaion record.txt show the best results every 50 epochs
6. the checkpoint.pth will be save in the directory checkpoint/ every 50 epochs

We can get 84% accuracy using EfficientNet.