Exploring knowledge distillation of unet for audio super resolution
torch==0.4.0torchsummary==1.5tqdm
First, we run the model on teacher mode and student mode：
usage: python train.py --mode MODE
optional arguments:
-h, --help show this help message and exit
--mode MODE Choose training mode, student or teacher
For example, to run the model on teacher mode, you would type:
python train.py --mode=teacher
After training the student model and the teacher model, we train the distillation model.
python train_dis.py
usage: python predict.py --mode MODE
optional arguments:
-h, --help show this help message and exit
--mode MODE Which model will be predicted? student, teacher or student_dis?
For example, to predict the model on teacher mode, you would type:
python predict.py --mode=teacher
get the lsd, snr matric of student, student_dis, teacher model
python get_matric.py