Open GreenLimeSia opened 2 years ago
Thanks for your attention to our work. We use schedule_sampling to help stabilize the training process. Schedule_sampling can help dynamically update the percentage of the ground truth images in the input images. For example, at the beginning of the training, most of the input images are the ground truth instead of the predictions. However, at the end of the training, most input images are predictions. The Schedule_sampling can help models progressively fit the training data and stabilize the training process.
Dear @ZhengChang467 : Thanks for your excellent codes. I have two questions about the function of schedule_sampling. 1) what is it used for? 2) why should we use this strategy?