search

adobe-research / SuperGaussian

ECCV 2024 SuperGaussian for generic 3D upsampling
https://supergaussian.github.io/
Other
27 stars 0 forks source link

Synthetic Dataset #5

Open SYXieee opened 4 days ago

SYXieee commented 4 days ago

How to run NeRF Synthetic Dataset?

MarcWangzhiru commented 4 days ago

I noticed in your post that you mentioned testing on the nerf synthetic dataset. Could you please make your test dataset publicly available?

yshen47 commented 3 days ago

For NerF synthetic, we use the same test set as prior work. Here are the detail described in the paper:

"On the Blender synthetic dataset, we follow the same experiment protocol for ×4 upsampling as prior work [32,49]. Specifically, we set the target upsampled resolution fixed at 800 × 800px and bilinearly downsample the target images to prepare low-res images to 200 × 200px. For fair comparisons against baselines, we also use Neural Radiance Field for 3D representation, instead of Gaussian Splattings. Results are evaluated on the original held-out test set in Blender synthetic dataset."