I tried to train PyMAF-X on AGORA dataset. I've produced corresponding .npz files using .pkls given by official. But I found it very different from other datasets, for AGORA has more than one individuals in each image and it has no lables such like 'scale' or 'center' (I don't know how to find the 2D coordinate of individuals for images). Could you please tell me more about your preprocessing with AGORA? How to train on it?
Currently I have two imagines. The first is to get segmented pictures for each individual for each image in AGORA. Then the data loading could be much easier. The second is to load the images and .pkls(.npzs) as is, and do some calculation to make it suitable for training. But this could cause troubles for dataloader.
At the end of the day I cannot find a proper way out. Begging for your help urgently.
Thanks for your excellent work.
I tried to train PyMAF-X on AGORA dataset. I've produced corresponding .npz files using .pkls given by official. But I found it very different from other datasets, for AGORA has more than one individuals in each image and it has no lables such like 'scale' or 'center' (I don't know how to find the 2D coordinate of individuals for images). Could you please tell me more about your preprocessing with AGORA? How to train on it?
Currently I have two imagines. The first is to get segmented pictures for each individual for each image in AGORA. Then the data loading could be much easier. The second is to load the images and .pkls(.npzs) as is, and do some calculation to make it suitable for training. But this could cause troubles for dataloader.
At the end of the day I cannot find a proper way out. Begging for your help urgently.
Thanks!