search

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
450 stars 72 forks source link

Question about how to generate the NOCS model #35

Closed ziyedy closed 3 years ago

ziyedy commented 4 years ago

Thanks for your sharing. I have a question that I can not find the source code which uses to generate the NOCS model (.obj file), and I really don't understand the detail how to design and generate the NOCS model. For example, I have no idea about how to generate a common camera model which can represent the category of Camera, and how can we get the NOCS value from this model. Looking foward to your reply.

hughw19 commented 3 years ago

I would suggest you go over the paper again. In NOCS, we don’t generate an exemplar model for a category. Instead, the network learns to predict the normalized coordinate, i.e. NOCS value, according to the true object shape. Basically, NOCS value only differs from the true coordinate by a similarity transformation (scale + 6D pose).

Best, He

On Mon, Nov 2, 2020 at 3:55 AM ziyedy notifications@github.com wrote:

Thanks for your sharing. I have a question that I can not find the source code which uses to generate the NOCS model (.obj file), and I really don't understand the detail how to design and generate the NOCS model. For example, I have no idea about how to generate a common camera model which can represent the category of Camera, and how can we get the NOCS value from this model. Looking foward to your reply.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/hughw19/NOCS_CVPR2019/issues/35, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEXSGH5AXKL6CCTPASV3PQ3SN2M2RANCNFSM4THL7DBQ .

WW-0 commented 3 years ago

请问作者，训练集里NOCS map的ground truth是怎么得到的

hughw19 commented 3 years ago

I just added an example code for using Blender to render a NOCS map. You may want to take a look.

hughw19 commented 3 years ago

请问作者，训练集里NOCS map的ground truth是怎么得到的

Please take a look at our paper. If your question is how ShapeNetCore model is normalized, the answer is they are already normalized as we described in the paper. Their original coordinates of each vertex, before you put the model into the scene, are NOCS ground truth.