metadata
license: other
license_name: realestate10k
license_link: https://google.github.io/realestate10k/
pretty_name: RealEstate10K test scenes (FineSplat / 2Xplat evaluation split)
task_categories:
- image-to-3d
tags:
- novel-view-synthesis
- gaussian-splatting
- realestate10k
RE10K-test
The 1,600 RealEstate10K test scenes used for the RE10K evaluation of FineSplat (and 2Xplat), at 640x360, one tar per scene.
tars/test_<id>.tar -> test_<id>/scene_meta.json (cameras, OpenCV convention, one entry per frame)
test_<id>/images/<frame>.png (640x360)
test_scenes.txt the 1,600 scene ids, in evaluation order
eval_ctx6_tgt8.json context / target frame indices per scene (6 context, 8 target views)
scene_meta.json lists, per frame, transform_matrix (camera-to-world), fl_x, fl_y, cx, cy,
w, h and the image path relative to the scene folder.
Usage
hf download HyebinKim/RE10K-test --repo-type dataset --local-dir re10k_test
cd re10k_test && for t in tars/*.tar; do tar xf "$t"; done # -> test_<id>/ folders
Then point data.data_path of the FineSplat RE10K eval configs at a text file listing the
test_<id>/scene_meta.json paths (test_scenes.txt gives the order).
License
RealEstate10K frames are extracted from YouTube videos and are redistributed here for research use only, under the terms of the original RealEstate10K release. If you use this data, cite RealEstate10K (Zhou et al., SIGGRAPH 2018).