Training with single dataset to generate table 1.

[heathentw]

For 3DPW as example, SOTA from recent years can reach about 70~90 on MPJPE, and the paper shows that smpler-x can only achieve 140.6 while train and test on 3DPW. Your code suggest that that MPJPE of 3DPW was computed with 14 points, which I assume was the same in other SOTA works, so your score can be compared with the SOTAs.

My question is that if the training results was far from expected standard, the story from table 1. will be questionable since the training part for each single dataset was not successfully completed, isn't it? Or there is some misunderstanding and the score is no comparable? Thank you.

[linjiangya]

same question here