search

JeongminB / E-D3DGS

[ECCV 2024] Official repository for "Per-Gaussian Embedding-Based Deformation for Deformable 3D Gaussian Splatting"
Other
109 stars 9 forks source link

How to downsample in training? #4

Closed zeshunzong closed 3 months ago

zeshunzong commented 3 months ago

Thanks for the nice work!

A quick question on downsampling. If I want to make the training faster by downsampling the training data, what is the most convenient way to achieve that?

seoha-kim commented 3 months ago

Thank you for your interest. We need this operation as well, so we'll check it and get back to you within a day. 😀

JeongminB commented 3 months ago

Hello!

You can downsample training images with -r or --resolution option (same as 3DGS). For example, if you add -r 2 argument to the command line, the training resolution will be 507 x 676 for the n3v dataset. (default is -r 1)

zeshunzong commented 3 months ago

by the way thanks for also fixing the loss_list index out of bound issue!

there is an annoying print in train.py line 151, it would be perfect if that is removed!