RPX: Robot Perception X

A real-world RGB-D benchmark for evaluating robot perception under embodied deployment conditions.

• Code: https://github.com/IRVLUTD/RPX

Dataset at a glance

Multi-object scenes (MOS)	100 (3 phases each: clutter / interaction / clean)
Single-object scenes (SOS)	220 (one 360° collection per object)
Total files	2,155,056
Total bytes	328.5 GB

Modality inventory

modality	files	bytes
cam_pose	185,000	102.3 MB
cam_pose_icp	16,574	312.9 MB
depth	185,000	24.5 GB
fisheye	370,000	102.3 GB
jpg	750	20.3 MB
mask_refinement	0	0 B
rgb	185,000	61.6 GB
sam2/_meta	1,685	562.9 KB
sam2/bbox_overlay	1,005	254.5 MB
sam2/contour_gt_masks	185,000	45.5 GB
sam2/dino_output	384,926	29.2 GB
sam2/mask_refinement	10,116	1.0 GB
sam2/masks	185,000	354.0 MB
sam2/masks_contour_with_hidden	75,000	23.5 GB
sam2/palette	185,000	413.2 MB
sam2/rgb_and_mask	185,000	39.4 GB

Quick start

pip install "rpx-benchmark[hub]"
hf auth login

from rpx_benchmark.dataset_hub import download_for_task

# Pull just RGB + masks for the Easy difficulty tier — never the whole repo.
res = download_for_task(task="segmentation", split="easy",
                          repo_id="itaykadosh/RPX")
print(res.local_dir, res.matched_scenes)

# Or from the CLI:
python -m rpx_benchmark.dataset_hub.cli download \
    --task segmentation --split easy \
    --repo-id itaykadosh/RPX

A subsequent call for a different task on the same split (e.g. relative_pose) reuses the cached RGB tars and only fetches the new modality (cam_pose) as the delta.

Repo layout

itaykadosh/RPX/
├── manifest/
│   ├── frames_v1.parquet     # per-frame metadata (always pulled, ~30 MB)
│   └── current.json          # default version per label modality
├── splits/
│   ├── scene_splits.json
│   ├── easy.txt  medium.txt  hard.txt
├── scenes/<scene_id>/<phase>/                     # MOS
│   ├── rgb.tar  depth.tar  fisheye.tar
│   └── labels/{cam_pose,masks,masks_aux,sam2_meta,vqa}/v1.tar
├── objects/<object_id>/0/                         # SOS
│   └── (same modality structure)
├── objects_meta/                                  # questionnaire dedup
│   ├── _index.json
│   └── <object_id>/questionnaire.json
└── README.md   ←  this file

Tasks

Multi-object (use a difficulty split)

recipe	inputs → labels
monocular_depth	['rgb'] → ['depth']
rgbd_segmentation	['depth', 'rgb'] → ['masks']
segmentation	['rgb'] → ['masks']
relative_pose	['rgb'] → ['cam_pose']
rgbd_relative_pose	['depth', 'rgb'] → ['cam_pose']
stereo_depth	['fisheye'] → ['depth']
object_tracking	['rgb'] → ['masks']
vqa	['rgb'] → ['questionnaire', 'vqa']

Single-object (no split — these are object templates)

recipe	inputs → labels
object_templates	['rgb'] → ['masks']
object_templates_rgbd	['depth', 'rgb'] → ['masks']
object_pose_library	['depth', 'rgb'] → ['cam_pose', 'masks']

Label versioning

Labels live at labels/<name>/v<N>.tar. Newer versions land at new paths; old versions stay reachable for reproducibility.

modality	current version
masks	v1
masks_aux	v1
sam2_meta	v1
cam_pose	v1

To pin to a specific version:

download_for_task(
    task="relative_pose", split="easy", repo_id="itaykadosh/RPX",
    label_versions={"cam_pose": "v1"},   # don't auto-upgrade to v2
)

Citation

@misc{rpx2026,
    title  = {RPX: Robot Perception X — A real-world RGB-D benchmark for
              embodied perception},
    author = {IRVL UT Dallas},
    year   = 2026,
    url    = {https://huggingface.co/datasets/itaykadosh/RPX},
}

License

Released under the cc-by-4.0 license.