Questions about "screen_point.npy"

[zhangwenniu]

Congratulations for acceptance of ICCV 2023! Thank you for your great contribution on transparent reconstruction, and thanks for your team work on a series of surface reconstruction following NeuS. I am interested in your work. For this NeTO reconstruction work, I am confused about the file of "screen_point.npy". I would like to know:

• What is this huge (as it is 2GB~3GB) file in each dataset?
• What role does it apply in transparent surface reconstruction?
• How do you get this file?

Furthermore, how do you construct your dataset? I am still not clear about manipulation details after reading your main paper and supplementary material. Thank you for your kind reply!

[zhangwenniu]

For example, in "hand" dataset: I noticed that screen_point.npy contains a ndarray with the shape (72, 1060, 1280, 3). I guess this correspond with each image (72 images) with 1060 x 1280 resolution picture, (3 dimension, xyz, but not sure this is world coordinate or camera coordinate). So I export each picture in ply file, and open it with cloud-compare software. It looks like:

• For one image index:

• For multiple image index:

What's their meaning?