Benchmark datasets

Preprocessed, canonically-laid-out copies of the datasets used by this repo's BenchmarkEvaluator (see docs/benchmark/evaluator_quickstart.md), packaged for direct download instead of re-running every scripts/download/ + scripts/preprocess/ step yourself.

Layout: each dataset scene is stored as one uncompressed <Task>/<dataset>/<scene>.tar (the raw tree is ~900k files). After downloading, extract everything in place with the bundled script:

huggingface-cli download Bluefish2333/benchmark --repo-type dataset --local-dir /path/to/benchmark_root
python /path/to/benchmark_root/extract_benchmark_archives.py /path/to/benchmark_root --delete

Every dataset folder below is redistributed under its own original source's license -- see that folder's own LICENSE_NOTICE.md. This repository's own license does not override those terms.

Included datasets

Task	Dataset	Scenes	License status	Source(s)	Data
MonoDepth	marigold	0	unconfirmed	NYU Depth V2, KITTI (depth selection / raw data), ETH3D, DIODE, ScanNet	packaged here (without scannet; see LICENSE_NOTICE.md)
MonoDepth	aerialmetric	0	unconfirmed	AerialMetric-Decoupled, AerialMetric-Oblique (test split), AerialMetric-Wild	packaged here
MonoDepth	hammer	0	safe	HAMMER (Highly Accurate Multi-Modal Dataset for DEnse 3D Scene Regression)	packaged here
MonoDepth	ibims1	0	safe	iBims-1 (independent Benchmark images and matched scans, version 1)	packaged here
MonoDepth	spring	0	safe	Spring	packaged here
MonoDepth	synth4k	0	unconfirmed	Synth4K (InfiniDepth game-captured 4K benchmark)	packaged here
VideoDepth	bonn	0	safe	Bonn RGB-D Dynamic Dataset	packaged here
VideoDepth	dycheck	0	safe	DyCheck (iPhone dataset, MegaSaM-distributed evaluation archive)	packaged here
VideoDepth	kitti	0	safe	KITTI (depth selection / raw data)	packaged here
SfM	eth3d	0	safe	ETH3D	packaged here
SfM	tanks_and_temples	0	safe	Tanks and Temples	packaged here
MVRecon	7scenes	0	safe	Microsoft 7-Scenes	packaged here
MVRecon	eth3d	0	safe	ETH3D	packaged here
MVRecon	tanks_and_temples	0	safe	Tanks and Temples	packaged here
MVRecon	scannetpp	0	safe	ScanNet++	packaged here
MVRecon	hiroom	0	safe	HiRoom (via DA3-BENCH)	packaged here
SLAM	7scenes	0	safe	Microsoft 7-Scenes	packaged here
SfM	scannetpp	0	safe	ScanNet++	packaged here
SfM	hiroom	0	safe	HiRoom (via DA3-BENCH)	packaged here
MVRecon	nrgbd	0	safe	Neural RGB-D Surface Reconstruction dataset	packaged here
SLAM	tum_dynamics	0	safe	TUM RGB-D SLAM Dataset (Dynamic Objects category)	packaged here
SLAM	oxford_spires	0	safe	The Oxford Spires Dataset	packaged here
SLAM	vbr	0	unconfirmed	VBR: A Vision Benchmark in Rome	packaged here
SLAM	kitti_odometry	0	safe	KITTI Odometry	packaged here
MonoDepth	spatialarena	19	safe	SpatialArena (the benchmark's own handheld 3-camera LiDAR rig captures)	packaged here
VideoDepth	spatialarena	15	safe	SpatialArena (the benchmark's own handheld 3-camera LiDAR rig captures)	packaged here
MVRecon	spatialarena	6	safe	SpatialArena (the benchmark's own handheld 3-camera LiDAR rig captures)	packaged here
SfM	spatialarena	7	safe	SpatialArena (the benchmark's own handheld 3-camera LiDAR rig captures)	packaged here
SLAM	spatialarena	8	safe	SpatialArena (the benchmark's own handheld 3-camera LiDAR rig captures)	packaged here

Excluded this run

Task	Dataset	Reason
VideoDepth	sintel	not present on disk
SfM	imc	not present on disk
MVRecon	dtu	not present on disk
SfM	dtu	not present on disk
SfM	texturepoorsfm	not present on disk