Question about dataset S2 (#seq)

NVlabs / dex-ycb-toolkit

A Python package that provides evaluation and visualization tools for the DexYCB dataset

https://dex-ycb.github.io

GNU General Public License v3.0

145 stars 24 forks source link

Question about dataset S2 (#seq) #4

Closed taeyeopl closed 3 years ago

taeyeopl commented 3 years ago

https://github.com/NVlabs/dex-ycb-toolkit/blob/e957432e0f81a477a897698340bbb69c05a3d9d6/dex_ycb_toolkit/dex_ycb.py#L157

In Tab 9 (main paper), the number of sequences (#seq) would be all 1,000 for train, Val, test instead of 750/125/125. As I understood, the number of sequences(#seq) is independent of the number of camera view points(#view). Have I missed something? Simple equation for #seq(1,000) = #subject_ind(10) * #sequnce_ind(100)

ychao-nvidia commented 3 years ago

You're right. This is an error. The number of sequences (#seq) for train/val/test should be 1,000/1,000/1,000 for S2 (unseen views) in the arXiv paper Tab. 9.

taeyeopl commented 3 years ago

Thanks for your reply!