about the joint order

[MengHao666]

First,thank u for such great work!
Now I have been confused about the joint order in the dataset.
In the ho3d_v2 dataset, you mention the joint order is different from that of MANO model.
And now I take part in the competetion Ho3d_v2 Codalab competition u provide, and I get the reults which have high error in joint loacation but much lower mesh error.
So I wonder if the joint order in the evaluation dataset is also different from that of original MANO model?
Hope you could provide some suggestion,thank u!

[shreyashampali]

hi,
The joint orderings follow the MANO model ordering even for the evaluation set as mentioned here. Could you please point me to the place where it is mentioned that the joint order is different from MANO model?

[MengHao666]

Thanks for the quick reply.
I'm sorry to confuse the code of jointsMapManoToSimple u provided and that of hassony's self.reorder_idxs.They may change the oreder to adapt to the differentiable Mano layer they proposed.

In my own understanding , u use jointsMapManoToSimple only for simple visualization ? Is that true?

Thanks u again!

[MengHao666]

Another question is that whether the order of samples in train.txt has been carefully shuffled to ensure uniform distribution.
Hope to get your suggestions, thanks!

[shreyashampali]

yes, jointsMapManoToSimple is only for visualization.
The samples in train.txt are just randomly shuffled. It does not guarantee uniform distribution

[MengHao666]

Ok,thanks a lot!