Closed syguan96 closed 3 months ago
Update: For Jitter, we should calculate on each sequence separately. For other metrics, we should calculate on the whole dataset.
When evaluating Transpose on DIP, you evaluate the mean error on each sequence and then average them. However, the sequence length is not equal, so you should save the errors of each sample and calculate the mean error on the whole dataset. Am I right?
Yes. The results in the paper can be reproduced using the current evaluation method. But definitely you can use the new averaging method.
When evaluating Transpose on DIP, you evaluate the mean error on each sequence and then average them. However, the sequence length is not equal, so you should save the errors of each sample and calculate the mean error on the whole dataset. Am I right?