设置sacle为0.3的时候报错

[SunyuKy]

在stage 0,当我没有为scale设置参数时是正常的，但是我想要更好的效果，所以设置bound 16，scale 0.3，确实取得了更高的psnr，但是在输出模型的时候报了以下错误，经过还原参数发型是设置了scale后出现的，请问你有出现过类似的情况吗？经过debug发现是在renderer.py 499行： vertices, triangles = mcubes.marching_cubes(-sigmas, 0) 执行后，vertices, triangles的值为空

‘’‘’
Traceback (most recent call last):
File "main.py", line 261, in
trainer.save_mesh(resolution=opt.mcubes_reso, decimate_target=opt.decimate_target, dataset=train_loader._data if opt.mesh_visibility_culling else None)
File "/root/nerf2mesh/nerf/utils.py", line 900, in save_mesh
self.model.export_stage0(save_path, resolution=resolution, decimate_target=decimate_target, dataset=dataset)
File "/root/miniconda3/lib/python3.8/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/root/nerf2mesh/nerf/renderer.py", line 509, in export_stage0
visibility_mask = self.mark_unseen_triangles(vertices, triangles, dataset.mvps, dataset.H, dataset.W).cpu().numpy()
File "/root/miniconda3/lib/python3.8/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/root/nerf2mesh/nerf/renderer.py", line 886, in mark_unseen_triangles
rast, _ = dr.rasterize(self.glctx, vertices_clip, triangles, (H, W)) # [1, H, W, 4]
File "/root/miniconda3/lib/python3.8/site-packages/nvdiffrast/torch/ops.py", line 310, in rasterize
return _rasterize_func.apply(glctx, pos, tri, resolution, ranges, grad_db, -1)
File "/root/miniconda3/lib/python3.8/site-packages/torch/autograd/function.py", line 506, in apply
return super().apply(*args, **kwargs) # type: ignore[misc]
File "/root/miniconda3/lib/python3.8/site-packages/nvdiffrast/torch/ops.py", line 248, in forward
out, out_db = _get_plugin().rasterize_fwd_cuda(raster_ctx.cpp_wrapper, pos, tri, resolution, ranges, peeling_idx)
RuntimeError: instance mode - pos must have shape [>0, >0, 4]
‘’‘’

[ashawkey]

@SunyuKy Hi, this indicates some outer cascade is totally empty, if you reduce scale, you may also consider reducing bound so it is not overly large and captures nothing.

[SunyuKy]

Thank you very much for your reply. After my test, it seems that the mesh can be output normally only when the scale is -1, and an error will be reported when the scale is other values. What version of pymcubes are you using? I'm wondering if it's a package version issue