What is DPG in training mode

[dekathomas]

Hello @Fang-Haoshu, what is the function of DPG when we train the model? This is the example of DPG value. I think it's not useful because when the epoch reach 140, it will reset the dataset and make the loss value increase

DPG_MILESTONE: 210
DPG_STEP:
- 20
- 40

This is the example of epoch in fast_421_res152_256x192.log

############# Starting Epoch 210 | LR: 1e-05 #############
Train-210 epoch | loss:0.00026830 | acc:0.8842
############# Starting Epoch 211 | LR: 0.001 #############
Train-211 epoch | loss:0.00030307 | acc:0.8559

Thank you

[dekathomas]

oke this one is data augmentation