search

jzhangbs / DTUeval-python

A fast python implementation of DTU MVS 2014 evaluation
MIT License
141 stars 18 forks source link

There is a big difference between mean_d2s and mean_s2d #7

Open wujinhu1999 opened 9 months ago

wujinhu1999 commented 9 months ago

Hello author, thank you for your great work! I would like to ask why my mean_d2s and mean_s2d are so different? Also, what happens when I get a visualize error here? Will it affect the final result? Looking forward to your reply! image

jzhangbs commented 9 months ago

Hi, the two metrics show your accuracy and completeness respectively. For example, if your result has a lot of holes, it will be revealed by the completeness score. Turning on visualization will not affect the result. From the visualization you can see which part of the point clouds produces large error.

wujinhu1999 commented 9 months ago

Hi, the two metrics show your accuracy and completeness respectively. For example, if your result has a lot of holes, it will be revealed by the completeness score. Turning on visualization will not affect the result. From the visualization you can see which part of the point clouds produces large error.

thank you for your reply! After I trained it with neus without mask, its completeness is quite high. But when I evaluate the CD distance, I need to clean it, which results in it being incomplete. before： image clean_mesh: image This will affect its completeness rating, right?

jzhangbs commented 9 months ago

yes. you can also check the error visualization