Ymx1025/FacadeTrack

Hurricane Helene Building Facade Dataset (Hugging Face export)

This folder contains a compact export of building facade images and per-building metadata from two street‑level data collection visits conducted around the time of Hurricane Helene. The images are rectified crops from panoramic video and are intended for tasks such as building condition/occupancy classification, change detection between visits, and related vision‑GIS experiments.

Contents

• Visit1_processed_images/ — JPEG facade crops for visit 1. Filenames use the pattern <ObjectId>.jpg and, when a second view exists, <ObjectId>_1.jpg.
• Visit2_processed_images/ — JPEG facade crops for visit 2. Same naming scheme as visit 1.
• Visit1_buildings.csv — Per‑building metadata for visit 1 (603 rows in this snapshot).
• Visit2_buildings.csv — Per‑building metadata for visit 2 (785 rows in this snapshot).

Metadata schema

Shared columns

• ObjectId — Integer building identifier. Matches the corresponding image filename stem.
• Center_Longitude, Center_Latitude — Building centroid in WGS84 (lon/lat).
• matched_file — Source video file identifier.
• frame_number — Frame index within the source video.
• vehicle_x, vehicle_y — Vehicle GPS at capture time (WGS84 lon/lat).
• PIN — Parcel/parcel‑level identifier (string).
• relative_time_ms — Milliseconds from the start of the source video to the matched frame.

Visit‑specific columns

• Status — Visit 1 only; free‑text status tag from preprocessing (e.g., existing).
• x, y — Visit 2 only; duplicate of centroid coordinates in some exports.
• orientation — Visit 2 only; viewing direction/yaw at capture time (degrees).

Mapping rows to images

• For a row with ObjectId = 443, the corresponding files are Visit1_processed_images/443.jpg (and possibly Visit1_processed_images/443_1.jpg) for visit 1; analogously for visit 2.
• Not every row is guaranteed to have both main and _1 images; prefer <ObjectId>.jpg and fall back to <ObjectId>_1.jpg if present.

Quick start (Python)

from pathlib import Path
import pandas as pd
from PIL import Image

root = Path('.')  # this folder
v = 1  # or 2
csv = pd.read_csv(root / f'Visit{v}_buildings.csv')
row = csv.sample(1, random_state=0).iloc[0]
obj = str(row['ObjectId'])

# Resolve image path (prefer main, then optional _1)
img_dir = root / f'Visit{v}_processed_images'
img_path = img_dir / f'{obj}.jpg'
if not img_path.exists():
    alt = img_dir / f'{obj}_1.jpg'
    if alt.exists():
        img_path = alt

print('Row info:', row.to_dict())
print('Image path:', img_path)
Image.open(img_path).show()

Notes and caveats

• Minor mismatches can exist between the CSV rows and available image files (e.g., only _1 present). Handle missing files defensively.
• Coordinates are in WGS84; join with parcel GIS by PIN after appropriate normalization.
• Images are rectified crops from 360° video; some scenes may include occlusions, vehicles, or motion blur.

License

MIT

Contact

For questions about this dataset or to request updates, please contact the project maintainers.