search

autonomousvision / monosdf

[NeurIPS'22] MonoSDF: Exploring Monocular Geometric Cues for Neural Implicit Surface Reconstruction
MIT License
573 stars 53 forks source link

Results of MLP and grids #37

Closed jzhzhang closed 2 years ago

jzhzhang commented 2 years ago

Thanks for sharing the code. We have tested the released pre-trained weights of both mlp and grids on scan 1.

But we found that the grids, which uses the instant-ngp, is worse than the simple mlp: scannet_grids.conf: image scannet_mlp.conf: image

I am not sure it's not an expected result of the paper. Can you give me some ideas about the results? Thanks in advance.

zhao-yiqun commented 2 years ago

You can refer to the section 4.1 and Table 2 of the paper for the experiment results between mlp and ngp, which illustrated that with both cues, ngp is worse than mlp.

jzhzhang commented 2 years ago

Thanks. Really appreciate your quick response.