Why should the reference pose be the first frame of the driven pose sequence?

[steven-kl]

Hi, during the inference, the reference pose is concatenated as the first frame with the target driven pose sequence. When I remove the reference pose, the results are quite bad and the generated first frame is still in the pose of the reference image. I'd like to know why is that?
Is that during the training, the first frame is used as the reference image?

[aleeyang]

same question

[JoJoistheBestOne]

it seems it is trained in this manner, but barely mentioned in paper....