YTxFSGAMERz/ARGUS_DATASET

🌍 ARGUS DATASET

Multi-Domain Global Landmark, Streetscape & Geospatial Intelligence Dataset

---

📌 Dataset Overview

ARGUS_DATASET is an open, research-grade geospatial intelligence (GEOINT), computer vision, and visual geolocation benchmark dataset. It provides verified, multi-angle landmark photography, panoramic street-level imagery, spatial index databases, and DCT perceptual hash trees across sovereign nations, territories, and municipalities worldwide.

The dataset powers visual reverse-geolocation engines, landmark retrieval models, spatial vision-language alignment (VLM), and cyber-physical sensor interrogation platforms.

📊 Aggregate Metrics

Dimension	Count / Metric	Technical Description
Total Landmarks	3,806	Synthesized from Wikidata SPARQL & OpenStreetMap Overpass
Landmarks with Valid Imagery	3,697 (97.1%)	Verified high-resolution imagery coverage
Total Visual Assets	88,800+	Cryptographically verified (SHA-256) & deduplicated binaries
Landmark Visual Assets	74,128+	Canonical perspective frames (Wikimedia Commons / GLDv2)
Street-Level Assets (SVI)	10,000+	Radial probes & NUS Global Streetscapes panoramic frames
Country Distribution	162 Nations	Global geographic dispersion across all continents
City Distribution	2,403 Municipalities	Distinct metropolitan and regional municipal zones
Surveillance Camera Grid	178,674 Nodes	Global CCTV sensor nodes with operator telemetry
ALPR Law Enforcement Links	4,253 Links	Inter-agency license plate reader data sharing links
Police Precinct Jurisdictions	272 Boundaries	GeoJSON multipolygon law enforcement precinct boundaries
Total Storage Footprint	~38.4 GB	High-resolution JPG payloads, SQLite DBs, and binary indices

---

🏗️ Repository Architecture

ARGUS_DATASET/
├── README.md                           # Standard Hugging Face Dataset Card & Documentation
├── .gitattributes                      # Git LFS configuration for binary payloads
│
├── ARGUS_DATASET/
│   ├── images/                         # 74,000+ Landmark visual assets categorized by ISO country
│   │   └── landmarks/
│   │       ├── AFG/                    # Afghanistan
│   │       ├── FRA/                    # France (e.g. Eiffel Tower, Louvre, Versailles)
│   │       ├── JPN/                    # Japan (e.g. Tokyo Tower, Fushimi Inari)
│   │       ├── USA/                    # United States (e.g. Statue of Liberty, Golden Gate)
│   │       └── ... [162 countries]
│   │
│   ├── places/
│   │   ├── places.csv                  # Master catalog of 3,806 landmarks with coordinates & IDs
│   │   ├── places.jsonl                # Line-delimited JSON representation of places
│   │   └── coverage/
│   │       ├── category_coverage.json  # Category-wise completeness and breakdown
│   │       └── country_coverage.json   # Country-wise landmark density distribution
│   │
│   ├── metadata/
│   │   ├── images.csv                  # Complete asset index: paths, dimensions, hashes, licensing
│   │   ├── images.jsonl                # Streamable JSONL image metadata records
│   │   ├── licenses.csv                # Legal taxonomy and redistribution rights mapping
│   │   └── sources.csv                 # Ingestion source endpoint provenance
│   │
│   ├── indexes/
│   │   ├── state_tracker.db            # Production SQLite database with spatial B-tree indexes
│   │   └── phash_bktree.index          # Serialized Burkhard-Keller tree for 64-bit DCT pHash lookup
│   │
│   └── reports/
│       ├── final_report.md             # Comprehensive audit, deduplication, and coverage report
│       └── license_audit.json          # Programmatic license verification log
│
├── data/
│   └── streetscapes/                   # NUS Global Streetscapes SVI dataset (10,000 observation frames)
│       ├── coords.csv                  # Geographic coordinates & headings
│       └── images/                     # 360-degree streetscape observation images
│
├── geodata/                            # 516 GeoJSON layers
│   ├── CAMERAS_WITH_NETWORK_DATA.geojson  # 178,674 surveillance camera nodes
│   ├── camera_networks.json               # Operator network cluster topologies
│   ├── police_precincts_usa.geojson       # US law enforcement precinct polygons
│   └── ...
│
└── geosent_chroma_db/                  # ChromaDB vector embedding index for semantic geosearch

---

📑 Data Schema & Field Definitions

1. places.csv (Landmark Catalog)

Field	Type	Description	Example
place_id	string	Unique UUID primary key (argus-place-...)	argus-place-c33a478a-1061-545f-8c39-b8a1989ce503
name	string	Canonical landmark name	Eiffel Tower
aliases	string	Semicolon-delimited aliases and local names	Tour Eiffel; Iron Lady
country	string	ISO 3166-1 alpha-3 sovereign country code	FRA
city	string	Municipality or urban district	Paris
category	string	Primary functional classification	tower
latitude	float	WGS84 decimal latitude	48.85837
longitude	float	WGS84 decimal longitude	2.294481
wikidata_id	string	Wikidata QID entity identifier	Q243
osm_id	string	OpenStreetMap relation/way/node identifier	relation/50138
wikipedia_url	string	Canonical Wikipedia encyclopedic reference	https://en.wikipedia.org/wiki/Eiffel_Tower
commons_url	string	Wikimedia Commons asset category gallery	https://commons.wikimedia.org/wiki/Category:Eiffel_Tower

Taxonomy Categories

palace, castle, fortress, tower, skyscraper, bridge, cathedral, place_of_worship, hindu_temple, mosque, museum, art_gallery, monument, archaeological_site, unesco_heritage, waterfall, mountain_peak, volcano, lighthouse, government, military.

---

2. images.csv (Asset Metadata)

Field	Type	Description
image_id	string	Unique UUID primary key (argus-img-...)
place_id	string	Foreign key referencing places.place_id
source	string	Ingestion source tier (Wikimedia Commons, Mapillary, NUS Streetscapes)
source_id	string	Upstream asset identifier or title
source_url	string	Direct HTTP upstream image URL
local_path	string	Relative path within repository to image binary
image_type	string	Asset modality (landmark, streetscape, aerial)
heading	float	Camera orientation heading in degrees ($0^\circ - 360^\circ$)
captured_at	string	Photographic timestamp or ISO date
width / height	int	Pixel dimensions
license	string	License designation (e.g. CC BY-SA 4.0, CC BY 2.0, CC0)
author	string	Original photographer or contributing organization
attribution	string	Full legal attribution statement
sha256	string	SHA-256 cryptographic checksum (exact byte deduplication)
phash	string	64-bit hexadecimal Discrete Cosine Transform (DCT) perceptual hash
file_size_bytes	int	Binary payload byte size

---

🚀 Quickstart & Usage

1. Load Metadata with Python & Pandas

import pandas as pd

# Load places catalog
places_url = "https://huggingface.co/datasets/YTxFSGAMERz/ARGUS_DATASET/raw/main/ARGUS_DATASET/places/places.csv"
df_places = pd.read_csv(places_url)
print(f"Loaded {len(df_places)} global landmarks.")
print(df_places[["name", "country", "city", "category", "latitude", "longitude"]].head())

# Load image metadata
images_url = "https://huggingface.co/datasets/YTxFSGAMERz/ARGUS_DATASET/raw/main/ARGUS_DATASET/metadata/images.csv"
df_images = pd.read_csv(images_url)
print(f"Indexed {len(df_images)} images across {df_images['place_id'].nunique()} landmarks.")

2. Download Image Asset via huggingface_hub

from huggingface_hub import hf_hub_download
from PIL import Image

# Download a specific landmark image
image_path = hf_hub_download(
    repo_id="YTxFSGAMERz/ARGUS_DATASET",
    repo_type="dataset",
    filename="ARGUS_DATASET/images/landmarks/FRA/argus-img-e41f57ff-ce30-562d-a174-3fc1145641fc.jpg"
)

img = Image.open(image_path)
print(f"Image Resolution: {img.size}")

3. Query Spatial Telemetry via SQLite (state_tracker.db)

import sqlite3
from huggingface_hub import hf_hub_download

db_path = hf_hub_download(
    repo_id="YTxFSGAMERz/ARGUS_DATASET",
    repo_type="dataset",
    filename="ARGUS_DATASET/indexes/state_tracker.db"
)

conn = sqlite3.connect(db_path)
cursor = conn.cursor()

# Query high-density landmark clusters in Italy
cursor.execute("""
    SELECT name, city, category, latitude, longitude 
    FROM places 
    WHERE country = 'ITA' 
    LIMIT 5
""")
for row in cursor.fetchall():
    print(row)

4. Visual Reverse Geolocation with Perceptual Hashing (pHash BK-Tree)

import pickle
from huggingface_hub import hf_hub_download

# Download pre-built Burkhard-Keller tree index
bktree_path = hf_hub_download(
    repo_id="YTxFSGAMERz/ARGUS_DATASET",
    repo_type="dataset",
    filename="ARGUS_DATASET/indexes/phash_bktree.index"
)

with open(bktree_path, "rb") as f:
    bktree = pickle.load(f)

# Query nearest visual matches within Hamming distance <= 10
# query_phash is an integer representation of a 64-bit DCT pHash
# results = bktree.query(query_phash, max_distance=10)

---

🔍 Quality Assurance & Deduplication Methodology

1. Two-Pass Deduplication Pipeline:
   • Pass 1 (Cryptographic SHA-256): Bit-exact duplicate payloads are eliminated immediately at download ingestion.
   • Pass 2 (Perceptual Hash BK-Tree): 64-bit DCT perceptual hashes are computed for every candidate frame. Near-identical photos ($d_H \le 10$) are retained only if camera azimuth differs by $\ge 30^\circ$ or acquisition date differs by $\ge 90\text{ days}$, ensuring viewpoint diversity (front, side, aerial, seasonal).
2. Geocoding Integrity: All landmark positions are cross-validated against Wikidata SPARQL coordinates and OpenStreetMap boundary polygons.
3. Format Normalization: Images are converted to standard RGB JPEG/PNG format with color space verification.

---

⚖️ Licensing & Attribution

• Landmark Images: Primarily distributed under Creative Commons Attribution-ShareAlike (CC BY-SA 4.0 / 3.0 / 2.0) and Public Domain / CC0. Detailed per-asset attribution, license URL, and original author details are strictly preserved in ARGUS_DATASET/metadata/images.csv.
• Streetscape Assets: Sourced from NUS Global Streetscapes and Mapillary v4 under CC BY-SA 4.0.
• Geodata & Boundaries: OpenStreetMap data is licensed under the Open Database License (ODbL).
• Indices & Code: Apache-2.0.

---

📬 Citation

If you use ARGUS_DATASET in your research, autonomous systems, or geospatial applications, please cite:

@dataset{argus_dataset_2026,
  author       = {YTxFSGAMERz},
  title        = {ARGUS DATASET: Multi-Domain Global Landmark, Streetscape, and Geospatial Intelligence Imagery},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/YTxFSGAMERz/ARGUS_DATASET}}
}