search

weiyithu / NerfingMVS

[ICCV 2021 Oral] NerfingMVS: Guided Optimization of Neural Radiance Fields for Indoor Multi-view Stereo
MIT License
435 stars 58 forks source link

Results quality of NeRF #11

Closed bennyguo closed 3 years ago

bennyguo commented 3 years ago

I downloaded the results you provided and found the results of NeRF do not match the ones shown in the paper. The NeRF results in the paper seem much worse. Could you please expain what is going on here? Thanks!

weiyithu commented 3 years ago

We do not provide the results of original NeRF. The results in nerf folder are generated by our method without post-filtering step, i.e., the fifth line in Table 3.

bennyguo commented 3 years ago

Thanks for your reply!

I wonder if you could share the training setting for the original NeRF on the ScanNet dataset (results in Fig.1). Did you use the same setting as that of the LLFF data? Thank you!

weiyithu commented 3 years ago

Yeah, it is similar to LLFF data. However, since scannet scene is 360 captured scene, you should add --no_ndc --spherify --lindisp flags.