Differences in validation metrics

[zhang-haojie]

I noticed that your validation metric on the HDTF dataset is 173, whereas another work EchoMimic your results shows a metric of 501. I would like to understand the specific details of the validation process. For instance, does ‘90% of the identity data’ refer to 90% of the data for each identity, or is it a random 90% sample across all identities? It also seems that the 188 clips you used from the HDTF dataset are significantly fewer than the total number of clips in the HDTF dataset