search

facebookresearch / localrf

An algorithm for reconstructing the radiance field of a large-scale scene from a single casually captured video.
MIT License
956 stars 62 forks source link

How to split train & test subsets for the Hike dataset #44

Closed liruilong940607 closed 8 months ago

liruilong940607 commented 8 months ago

Thanks for the great work!

I'm trying to do some comparison with your work on the Hike dataset. To do that I need to know how the train & test sets are constructed from all images.

But after going thru the code it seems like you are not excluding the test set from train set? I.e. is it true that the train set you are using contains all the images?

https://github.com/facebookresearch/localrf/blob/b1b14e72ef8c8298fcd41e5edf676bb81c758fa8/localTensoRF/dataLoader/localrf_dataset.py#L77-L96

Best, Ruilong

ameuleman commented 8 months ago

Hi,

train_dataset loads all images but only samples from the training set when optimizing the radiance fields https://github.com/facebookresearch/localrf/blob/b1b14e72ef8c8298fcd41e5edf676bb81c758fa8/localTensoRF/dataLoader/localrf_dataset.py#L273-L281 https://github.com/facebookresearch/localrf/blob/b1b14e72ef8c8298fcd41e5edf676bb81c758fa8/localTensoRF/train.py#L432-L438

liruilong940607 commented 8 months ago

Oh cool thanks for the clarification! This makes a lot of sense now!