{ "previewOnly": true, "rows": 96, "bytes": 3306503, "features": { "sample_id": { "dtype": "string", "_type": "Value" }, "split": { "dtype": "string", "_type": "Value" }, "variant": { "dtype": "string", "_type": "Value" }, "pair_key": { "dtype": "string", "_type": "Value" }, "session": { "dtype": "string", "_type": "Value" }, "frame_idx": { "dtype": "int64", "_type": "Value" }, "global_frame_idx": { "dtype": "int64", "_type": "Value" }, "source_archive": { "dtype": "string", "_type": "Value" }, "source_member": { "dtype": "string", "_type": "Value" }, "source_mat_sha256": { "dtype": "string", "_type": "Value" }, "unit": { "dtype": "string", "_type": "Value" }, "point_count": { "dtype": "int64", "_type": "Value" }, "rotation_degrees": { "dtype": "float64", "_type": "Value" }, "sensor_names": { "feature": { "dtype": "string", "_type": "Value" }, "length": 6, "_type": "Sequence" }, "sensor_xyz_mm": { "feature": { "feature": { "dtype": "float32", "_type": "Value" }, "length": 3, "_type": "Sequence" }, "length": 6, "_type": "Sequence" }, "sensor_heatmap_peak_uv": { "feature": { "feature": { "dtype": "int64", "_type": "Value" }, "length": 2, "_type": "Sequence" }, "length": 6, "_type": "Sequence" }, "sensor_projected_uv": { "feature": { "feature": { "dtype": "float32", "_type": "Value" }, "length": 2, "_type": "Sequence" }, "length": 6, "_type": "Sequence" }, "sensor_rotations_rowmajor": { "feature": { "feature": { "dtype": "float32", "_type": "Value" }, "length": 9, "_type": "Sequence" }, "length": 6, "_type": "Sequence" }, "bbox": { "feature": { "dtype": "float32", "_type": "Value" }, "length": 4, "_type": "Sequence" }, "depth_display_min_mm": { "dtype": "float32", "_type": "Value" }, "depth_display_max_mm": { "dtype": "float32", "_type": "Value" }, "depth": { "decode": false, "_type": "Image" }, "infrared": { "decode": false, "_type": "Image" }, "labels": { "decode": false, "_type": "Image" }, "preview_note": { "dtype": "string", "_type": "Value" } }, "description": "HandNet derived 96-row visualization preview. Deterministic stratified subset: 24 test original plus 24 matched perturbed, 16 validation original plus 16 matched derotated, 16 training derotated. Depth and IR are contrast-rendered preview PNGs; labels use a fixed palette. Source archives retain raw MATLAB arrays and all original data. Six sensors: thumb, index, middle, ring, little, palm. Camera coordinates are in millimetres, +X right, +Y down, +Z away.", "sourceManifest": "space-data/samples.json", "validation": "96 rows, Hugging Face schema metadata, Dataset.from_parquet round trip and first embedded image bytes verified." }