From 54bdab0aa5e32dc60530f3c1cb834b23214d1cb1 Mon Sep 17 00:00:00 2001 From: paulbzm Date: Tue, 17 Sep 2024 12:45:27 +0200 Subject: [PATCH] update readme --- README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index f9106b7..bf30a17 100644 --- a/README.md +++ b/README.md @@ -71,13 +71,13 @@ Include up to two consistent submethod names for your results across all files t You can view an example [here](https://github.com/fraunhoferhhi/Self-Organizing-Gaussians/tree/main/results) -### Caution: 3DGS testing conventions +### Important: 3DGS testing conventions Authors are required to adhere to the testing conventions established in the original [3DGS project](https://github.com/graphdeco-inria/gaussian-splatting). Specifically, this includes: - Using all 9 scenes from the MipNeRF360 dataset, including "flowers" and "treehill". - Evaluating images at full resolution up to a maximum width of 1600px. Larger test images should be downscaled to a width of 1600px (which is only applicable for MipNeRF360). -- For the 3 COLMAP datasets (Tanks and Temples, Deep Blending, MipNeRF360), use every 8th image for testing. Concretly, the test images are those where idx % 8 == 0. +- For the 3 COLMAP datasets (Tanks and Temples, Deep Blending, MipNeRF360), use every 8th image for testing. Concretly, the test images are those where ```idx % 8 == 0```. - For the Blender dataset (SyntheticNeRF), follow the predefined train/eval split.