Datasets:

facebook
/

xrstereo

Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 84, in _split_generators
                  raise ValueError(
                  ...<2 lines>...
                  )
              ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

XRStereo

Synthetic stereo video dataset introduced with StereoSplat: Metric-Scale Novel View Synthesis via Stereo-Grounded Gaussian Splatting (NeurIPS 2026).

Paper | Project Page | Code

Sequences of rectified stereo pairs rendered from moving cameras in static scenes, with dense disparity for both views, intrinsics, baseline and camera poses. This release is the StereoSplat training split.


Sequences	17,922 (`train` split)
Stereo frames per sequence	5
Resolution	1024 x 1024, RGB, 8 bit
Horizontal field of view	50 to 130 degrees
Stereo baseline	40 to 120 mm
Size	372.8 GB in 72 tar shards

Download

hf download facebook/xrstereo --repo-type dataset --local-dir xrstereo
cd xrstereo
for t in data/*.tar; do tar -xf "$t"; done

manifest.json lists the sequences, size and SHA-256 of each shard.

Structure

sequences/
  seq_000000/
    calibration.json     {"f", "cx", "cy", "baseline_m"}
    poses_left.npy       (5, 4, 4) float64, world-from-camera pose of the left camera
    frame_0000.npz       ... frame_0004.npz, in temporal order
  ...
splits/train.txt         one sequence id per line

Each frame_XXXX.npz holds:

Key	dtype, shape	Content
`image_left`, `image_right`	uint8, (1024, 1024, 3)	RGB images
`disparity_left`	float16, (1024, 1024)	left pixel x matches right pixel x - d
`disparity_right`	float16, (1024, 1024)	right pixel x matches left pixel x + d

Disparity <= 0 is invalid (e.g. sky). Disparity is measured through transparent surfaces such as glass.

Cameras

Both cameras share the intrinsics K = [[f, 0, cx], [0, f, cy], [0, 0, 1]] (pixels), constant over the sequence.
The right camera is the left camera translated by baseline_m metres along its x axis.
Metric depth: depth = f * baseline_m / disparity.
poses_left.npy is in metres, in Blender camera axes (x right, y up, z backward). For OpenCV axes, multiply on the right by diag(1, -1, -1, 1).

import json
import numpy as np

seq = "sequences/seq_000000"
calib = json.load(open(f"{seq}/calibration.json"))
c2w_left = np.load(f"{seq}/poses_left.npy") @ np.diag([1.0, -1.0, -1.0, 1.0])  # OpenCV
with np.load(f"{seq}/frame_0000.npz") as z:
    disp = z["disparity_left"].astype(np.float32)
depth = np.where(disp > 0, calib["f"] * calib["baseline_m"] / np.maximum(disp, 1e-8), 0.0)

License

CC BY-NC 4.0, see LICENSE. Non-commercial use only.

Citation

@inproceedings{yugay2026stereosplat,
  title     = {StereoSplat: Metric-Scale Novel View Synthesis via Stereo-Grounded Gaussian Splatting},
  author    = {Yugay, Vladimir and Rozumny, Denis and Gevers, Theo and Vansteenkiste, Elias and Oswald, Martin R. and Luvizon, Diogo C.},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2026}
}

Downloads last month: 201

Total file size:

373 GB

Paper for facebook/xrstereo

StereoSplat: Metric-Scale Novel View Synthesis via Stereo-Grounded Gaussian Splatting

Paper • 2610.35902 • Published Sep 29