Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
image
image
End of preview. Expand in Data Studio

IMCUI Examples

Curated image subsets used by the example galleries of Image Matching WebUI.

This repository contains only the images referenced by the WebUI galleries. Each gallery downloads its own subset with snapshot_download(..., allow_patterns=...); nothing else is fetched.

Legacy repository: Realcat/imcui_datasets is kept unchanged for older releases. It still hosts the original parquet pack (Sacre Coeur, WxBS, and line images). New releases use this repository. HardMatch remains in Realcat/hardmatch.

Layout

sacre_coeur/
  mapping/            10 images
  mapping_rot/        70 images (45, 90, 135, 180, 225, 270, 315 degrees)
  mapping_scale/      70 images (0.2, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9)
fisheye/
  construction/
    pillar_turn/      A.jpg, B.jpg
    corridor/         A.jpg, B.jpg
    glazed_walkway/   A.jpg, B.jpg
  street/
    facade_row/       A.jpg, B.jpg
    avenue/           A.jpg, B.jpg
    intersection/     A.jpg, B.jpg
Subset Files Size Gallery Downloaded by default
sacre_coeur/ 150 14.7 MB Classic 70 files / 9.4 MB
fisheye/ 12 1.8 MB Fisheye 12 files / 1.8 MB

Sizes are decimal MB. The WebUI downloads only the files it references, not the whole repository.

Galleries

Classic: Sacre Coeur

Ten base images, seven rotations per image, and seven scales per image. The WebUI samples five unmodified pairs, five rotated pairs, and five scaled pairs at runtime, so the download includes every candidate the sampler may pick: the ten base images, the five rotations used by the UI (45, 90, 180, 225, and 270 degrees), and the 0.5 scale. The 135/315-degree rotations and the six other scales (0.2, 0.4, 0.6, 0.7, 0.8, and 0.9) are kept for future use but are not downloaded.

Source: Image Matching Challenge 2021, originally from YFCC100M. See sacre_coeur/README.md.

Fisheye

Six pairs of raw fisheye frames. No undistortion is applied, so homography and fundamental-matrix estimates are approximate: fisheye projection bends epipolar lines. The pairs are a stress test for feature matchers, not a calibrated stereo benchmark.

Pair Frames (A to B) Camera Source
construction/pillar_turn 10000000000000.jpg to 10001501519000.jpg Hilti cam0 Hilti-Trimble SLAM Challenge 2026
construction/corridor 10048982292000.jpg to 10050316938000.jpg Hilti cam0 Hilti-Trimble SLAM Challenge 2026
construction/glazed_walkway 10073473404000.jpg to 10074808108000.jpg Hilti cam0 Hilti-Trimble SLAM Challenge 2026
street/facade_row 0306.png to 0312.png sunset cam1 Sunny / Cloudy / Sunset
street/avenue 0455.png to 0475.png sunset cam1 Sunny / Cloudy / Sunset
street/intersection 0909.png to 0929.png sunset cam1 Sunny / Cloudy / Sunset

The Hilti frames are equidistant fisheye, originally 1472x1440. They are resized to a 1024 px long side and saved as JPEG quality 90. The sunset frames are ocam fisheye, natively 800x768, saved as JPEG quality 90. These are display copies; use the original datasets for calibration experiments.

Download only one gallery

from huggingface_hub import snapshot_download

# Classic gallery: 70 files, about 9.4 MB
snapshot_download(
    "Realcat/imcui_examples",
    repo_type="dataset",
    allow_patterns=[
        "sacre_coeur/mapping/*.jpg",
        "sacre_coeur/mapping_rot/*_rot45.jpg",
        "sacre_coeur/mapping_rot/*_rot90.jpg",
        "sacre_coeur/mapping_rot/*_rot180.jpg",
        "sacre_coeur/mapping_rot/*_rot225.jpg",
        "sacre_coeur/mapping_rot/*_rot270.jpg",
        "sacre_coeur/mapping_scale/*_scale0.5.jpg",
    ],
    local_dir="examples",
)

# Fisheye gallery: 12 files, about 1.8 MB
snapshot_download(
    "Realcat/imcui_examples",
    repo_type="dataset",
    allow_patterns=["fisheye/**"],
    local_dir="examples",
)

The WebUI uses the same patterns; it never downloads the whole repository.

Licensing and attribution

This repository mixes sources with different terms. Check each subset before use.

  • Sacre Coeur - images from the Image Matching Challenge 2021, originally from YFCC100M. The original images retain their YFCC100M licenses. See sacre_coeur/README.md for links.
  • Hilti-Trimble SLAM Challenge 2026 - the three construction/ pairs. Licensed under CC BY-NC-SA 3.0 according to the official dataset page. Attribution: Hilti-Trimble SLAM Challenge 2026. Non-commercial use only; derivative works must be shared under the same license.
  • Sunny / Cloudy / Sunset dataset by Changhee Won - the three street/ pairs. No explicit license is declared in the released dataset. They are included here for research and demonstration; contact the dataset author for other uses.

If you are a rights holder and want a subset removed, please open an issue in the WebUI repository.

Related datasets

Downloads last month
-