From my understanding, the poses of testing images are not optimised when training the BARF, so when rendering the raw pose, the rendered image cannot be compared with the testing image, cause the training poses are optimised, the testing poses are not, they are sampled from different angles. So I was wondering how you evaluate the quantitive metric at the eval stage.
Any help would be highly appreciated, thanks in advance.
Thanks for your great work!
From my understanding, the poses of testing images are not optimised when training the BARF, so when rendering the raw pose, the rendered image cannot be compared with the testing image, cause the training poses are optimised, the testing poses are not, they are sampled from different angles. So I was wondering how you evaluate the quantitive metric at the eval stage.
Any help would be highly appreciated, thanks in advance.