Dataset layout

data/generated/<dataset>/
|-- dataset_manifest.json
`-- scene_001_<label>/
    |-- scene_info.json
    |-- provenance.json
    |-- source_buildings.json
    |-- scene.xml
    |-- mesh/
    |-- terrain/
    |-- measurement_surfaces/
    |-- vox_slices/
    |-- vox_depth/
    |-- 2D_Building_Height_Map.npy
    `-- cases/
        `-- case_0001_<antenna>_<n>tx_<frequency>ghz/
            |-- case_info.json
            |-- arrays/
            |   |-- path_gain/path_gain.npz
            |   |-- rss/rss.npz
            |   `-- sinr/sinr.npz
            `-- images/
                |-- path_gain_db.png, path_gain_db_masked.png
                |-- rss_db.png, rss_db_masked.png
                |-- sinr.png, sinr_masked.png
                `-- association.png

Manifest

dataset_manifest.json holds the validated request, the expanded plan with every case’s seed and transmitter parameters, the software version, the resume contract, per-scene status, and a generation_runs list with the initial, generated and failed case counts of each run.

Scene directory

Each scene directory is a complete compiled scene (Scene layout) plus the derived assets that its case space needs: one measurement surface per requested cell size and receiver height, voxel slices and depth at the requested pitch. With include_terrain=True the terrain/ arrays are written for every scene and are reloaded for resume without a new provider query.

Case directory

case_info.json is mandatory and records the resolved simulation request, the solver details and warnings, the data provenance of buildings, terrain and surfaces, and the artifact list. The arrays are compressed NPZ files containing the linear float32 metric array with shape (n_tx, rows, columns), rows south to north. PNGs are north-up, one pixel per cell, with transparent no-data and no interpolation; the _masked variants hide building pixels. association.png is rendered from the discrete serving-transmitter labels.

Metric	Array unit	Image scale
path_gain	linear ratio	dB
rss	W	dBm
sinr	linear ratio	dB
association	integer transmitter index	categorical

Reading a dataset

import json
import numpy as np
from pathlib import Path

case = Path("data/generated/my-radio-dataset/scene_001_amsterdam/cases/case_0001_sector_2tx_3.5ghz")
info = json.loads((case / "case_info.json").read_text())
rss_w = np.load(case / "arrays/rss/rss.npz")["rss"]          # (n_tx, rows, columns)
best_dbm = 10 * np.log10(rss_w.max(axis=0)) + 30              # strongest transmitter per cell

scene = case.parents[1]
local_elevation = np.load(scene / "terrain/local_elevation_m.npy")
surface_classes = np.load(scene / "terrain/surface_classes.npy")

Elevation arrays are grid nodes and surface classes are cells, so their shapes differ by one in each dimension. Use the coordinates stored with each product when aligning them with the north-up radio images.