trade_vision_dataset

TradeVision: Hierarchical Physical Business & Multimodal Retail Provenance Dataset

This dataset is continuously seeded from OpenStreetMap, matched to Google Place IDs, harvested for temporal store photos, and enriched with zero-shot computer vision using Hugging Face Hub native pipelines.

Dataset Structure

The dataset is partitioned into three relational subsets loadable via Hugging Face datasets:

from datasets import load_dataset

# 1. Load Canonical Businesses (Hierarchical Admin 0-3 with OSM & Google Place IDs)
registry = load_dataset("drksci/trade_vision_dataset", "registry")

# 2. Load Time-Stamped Photos & Dense SigLIP Embeddings
photos = load_dataset("drksci/trade_vision_dataset", "photos")

# 3. Load 3-Level Retail Taxonomy Detections
detections = load_dataset("drksci/trade_vision_dataset", "detections")

1. registry Subset

Contains every verified physical retail location with multi-authority cross-references: - business_id: Deterministic unique identifier - name, brand, category - admin_level_0 (Country) -> admin_level_1 (State) -> admin_level_2 (City/LGA) -> admin_level_3 (Suburb) - osm_id & osm_url: Canonical OpenStreetMap reference - google_place_id & google_maps_url: Google Maps live location - latitude, longitude

2. photos Subset

Contains full temporal provenance and visual semantic context: - photo_id: SHA-256 content hash - raw_source_uri & image_url_highres: Original extraction payload and uncompressed image - upload_date_raw & upload_timestamp_approx: Visible upload timestamp (e.g. Dec 2025, Mar 2026) - scene_coarse & scene_functional_zone: Zero-shot environmental classification - image_embedding: Normalized 768-dimensional dense vector (google/siglip-base-patch16-224)

3. detections Subset

Contains hierarchical retail merchandising objects detected via open-vocabulary vision: - taxonomy_l1 (Super-Category: signage, fixtures_and_merchandising, point_of_sale) - taxonomy_l2 (Category: digital_signage, static_signage, display_fixtures, etc.) - taxonomy_l3 (Leaf Artifact: wall_mounted_display, promotional_poster, gondola_shelving, etc.) - detected_label, confidence, bbox_xyxy, placement_estimate

Continuous Updates & Ingestion

This dataset updates automatically via background GitHub Actions runners and Hugging Face Serverless APIs.