OmniSearch: Annotated CV Detection Visualizations (Physical Scale Verification)

Annotated visualization export accompanying the OmniSearch synthetic survivor-detection datasets (see companion record DOI 10.5281/zenodo.21226010). Every frame from the drone, UGV (front/mast), synthetic-thermal, and real thermal (HIT-UAV) training sets is re-rendered with: Bounding boxes for every labeled survivor, with pixel size AND real-world size in metres (px × ground-sample-distance) Color-coded hard-negative decoy boxes (orange = vehicle, red = animal, magenta = colorful object) drawn from generation metadata, unlabeled A header banner with GSD (m/px), altitude, camera mode (nadir/oblique/UGV), and survivor/decoy counts per frame A scale bar calibrated to the frame's terrain resolution Purpose: let a reviewer visually confirm that every composited object (survivor and decoy) is physically plausible relative to the terrain scale, without re-running any generation code. Frames from the real HIT-UAV thermal set are flagged 'NO SCALE' since that public dataset does not publish per-frame camera intrinsics. Packaged as one zip per dataset (survivor, survivor_6k, survivor_naip, thermal, thermal_real, ugv).