Thanks a lot for open-sourcing the code for PointCLIPv2. I have a few questions regarding the execution of the model to test open-world understanding.
I see that there are two main contributions of v2 compared to PointCLIP, i.e (1) realistic projection and (2) prompting GPT-3 for 3D-specific text. For the second part, there are weights provided to decide the best prompt.
How can these weights be calculated for arbitrary point clouds (or other datasets) apart from ModelNet40 and ScanObjectNN? Is there a script that can be used for computation?
Hi!
Thanks a lot for open-sourcing the code for PointCLIPv2. I have a few questions regarding the execution of the model to test open-world understanding.
I see that there are two main contributions of v2 compared to PointCLIP, i.e (1) realistic projection and (2) prompting GPT-3 for 3D-specific text. For the second part, there are weights provided to decide the best prompt.
How can these weights be calculated for arbitrary point clouds (or other datasets) apart from ModelNet40 and ScanObjectNN? Is there a script that can be used for computation?
Thanks in advance!