the result's resolution

[duzhenjiang113]

There are three different resolutions written in your training code, so could your network produce results with several different resolutions? thanks

[ChrisWu1997]

Yes, our network can produce results at different resolutions (specify this arg). This property comes from the implicit decoder that we used, but has nothing to do with using three different resolutions in training. Multi-scale progressive training strategy is just to speed up the training and convergence (see Sec. 3.1 of IM-NET paper).

[duzhenjiang113]

thanks

[CRISZJ]

i have a question is that can we get the each part result with 32^3 resolution. thanks

[ChrisWu1997]

Yes, we can. The code does not provide a direct API for this so you need to modify the code a little bit to achieve this:

After this for loop (before transform into global coordinates in Line 202), you will get (points, values) pairs for each part of its own local space. Then apply function sdf2voxel to convert each part to voxel representation, and the desired resolution is passed as an argument.

[CRISZJ]

Yes, we can. The code does not provide a direct API for this so you need to modify the code a little bit to achieve this:

After this for loop (before transform into global coordinates in Line 202), you will get (points, values) pairs for each part of its own local space. Then apply function sdf2voxel to convert each part to voxel representation, and the desired resolution is passed as an argument.

ok，thank you