search

Colin97 / OpenShape_code

official code of “OpenShape: Scaling Up 3D Shape Representation Towards Open-World Understanding”
https://colin97.github.io/OpenShape/
Apache License 2.0
224 stars 15 forks source link

About Custom Mesh Data Preprocessing #16

Closed kizum closed 7 months ago

kizum commented 9 months ago

Great work! Thanks for your kindly sharing this project with us!

Based on your work, I want to run some shape retrieval demo on my custom mesh data (just assume they are some .obj files). I have noticed that the whole project is based on the point cloud representation. As the paper and repo mentioned, I think I should sample the mesh file into a point clouds file containing 10000 points and align the gravity axis of the mesh.

However, I am not very sure about the preprocessing details. Could you please give me some hints or provide some example codes (maybe some codes when you process the .glb file in the objaverse)?

Thanks!

Junking1 commented 9 months ago

same question +1

Colin97 commented 9 months ago

Please refer to the code here.

kizum commented 9 months ago

Please refer to the code here.

Thanks a lot!

kizum commented 8 months ago

Please refer to the code here.

Hi, sorry for bother again. In the git repo and this model page: https://huggingface.co/OpenShape/openshape-demo-support, I find that you have mentioned that the gravity axis for G14 model is z-axis, and for B32/L14 is y-axis.

In the src/example.py file, the script evaluate the similarity between the demo/owl.ply file and some images/texts. If I have no mistake, the owl.ply file should be a y-gravity-axis point cloud file, like showed in meshlab:

image

And the output similarity score is like: image

So, in my mind, if I convert the owl.py into a z-gravity-axis point cloud file, the score should be higher and better. After conversion, the point cloud looks like: image

However, when I evaluate the similarity on the new point clouds, the score get lower. image

Is there anything wrong, or this is just right?

Colin97 commented 7 months ago

Hi, sorry for the late reply.

In our data loader, we already switched the gravity axis by default (see here). So there is no need to change the ply file

kizum commented 7 months ago

Hi, sorry for the late reply.

In our data loader, we already switched the gravity axis by default (see here). So there is no need to change the ply file

Oh, I missed that. So the performance makes sense here. Thanks for your reply!