ZhoutongZhang
ZhoutongZhang
We use distance function instead of signed distance function, since most of the meshes are not watertight. I'm not sure if I can trace down the exact code for df...
I don't think we currently support RTX cards. I'll add a todo tag for this.
@andywu1018, currently the texture map is implicit. The only way to generate texture map is to generate multiple views of an object and perform back projection. We are currently working...
I think this sounds like your cuda & pytorch & gcc versions are in a weird combination. Does the dock image work for you?
1. the thresholding level is different since the data stored is distance function, which is stored as the distance between the voxel grid's center to the nearest surface. We convert...
Hi, pytorch has changed its ffi interface and sorry we currently do not have the bandwidth to do a full upgrade. I'll see if I can find time to do...
You need to have a repository of 3D shapes for the specific class, and convert them to voxels. Secondly, you need to have a repository of images with mask, and...
Please share your pytorch version. This seems to be a mismatch due to newer version of pytorch.
re 1: The the loaded mask during training is further preprocessed by this function here: https://github.com/xiumingzhang/GenRe-ShapeHD/blob/ee42add2707de509b5914ab444ae91b832f75981/models/marrnetbase.py#L104, which converts it into a single channel mask, so there shouldn't be any difference...
hi, I think that sounds like a pytorch version issue. Newer version of pytorch does not support the cffi interface anymore.