hughw19 / NOCS_CVPR2019

[CVPR2019 Oral] Normalized Object Coordinate Space for Category-Level 6D Object Pose and Size Estimation on Python3, Tensorflow, and Keras
https://geometry.stanford.edu/projects/NOCS_CVPR2019/
Other
453 stars 72 forks source link

Isn't there a single NOCS for each category? #60

Closed N-NEJATISHAHIDIN closed 3 years ago

N-NEJATISHAHIDIN commented 3 years ago

Such an excellent paper and idea! I had a question regarding how you overlay different canonicalized CAD models to get the NOCS? As I went through the paper, I came up with the idea that the ground truth NOCS for each instance is different (the space is the same, but the NOCS map for 2 instances with the same pose is different). So you did not overlay the canonicalized CAD model to build the NOCS. But, you mention in the paper abstract "a shared canonical representation for all possible object instances within a category" I was confused about this term. Thanks!

N-NEJATISHAHIDIN commented 3 years ago

https://github.com/hughw19/NOCS_CVPR2019/issues/18#issue-556635373