mothbot-detection-json
All format examples · Machine-readable schema · Convert outputs to COCO
Illustrative detection-stage annotation constructed independently of probe evidence. Four pixel vertices describe an oriented box; direction uses OpenCV degrees. difficult is a string. version is the detector name. Descriptions and identifiers may contain nonempty strings.
Example
Illustrative values, not predictions from a model. Unless stated otherwise, the image is 200 pixels wide and 100 pixels high.
{
"description": "",
"flags": {},
"imageData": null,
"imageHeight": 100,
"imagePath": "frame.jpg",
"imageWidth": 200,
"shapes": [
{
"attributes": {},
"confidence_detection": 0.8,
"description": "",
"detector_bot": "MBD-1-1",
"difficult": "false",
"direction": 90,
"flags": {},
"group_id": null,
"identifier_bot": "",
"identifier_human": "",
"kie_linking": [],
"label": "creature",
"patch_path": "frame_0_MBD-1-1.jpg",
"points": [
[10, 20],
[50, 20],
[50, 60],
[10, 60]
],
"score": 0.8,
"shape_type": "rotation",
"timestamp_detection": "2026-09-30__12_00_00_(+0000)"
}
],
"version": "MBD-1-1"
}Convert to a standard format
Use Converter.convert with source="mothbot-detection-json" and target="coco". Your application supplies the predictions and any required image context or producer settings. The adapter validates whether the output contains the information the target needs.
General conversion example and required context
How to interpret it
Scope: Fresh detection-stage output for the PyTorch OBB path. Later identification/manual editing/tracking can change labels and add fields; this profile does not describe those later stages or the separate ONNX writer. The null-only group_id and imageData describe this writer, which does not populate grouping or embed image bytes. Supporting those features requires a separately evidenced extension. The rotation literal and four point positions define this OBB representation, rather than defaults for arbitrary annotation shapes.
Coordinates: points contains four [x,y] vertices in original-image pixels. direction is cv2.minAreaRect angle in degrees, not the YOLO internal rotation in radians.
Confidence: score and confidence_detection duplicate the OBB confidence in [0,1] when first written. The detector initializes label to creature and does not export model class IDs; label contents are not constrained to that initial value.
Identity: version and detector_bot contain the model name, not a schema version. imagePath is the supplied image path. patch_path is a precomputed filename when thumbnails are enabled, otherwise an empty string; thumbnail existence is a separate step.
Timestamp: timestamp_detection uses local YYYY-MM-DD__HH_MM_SS_(+HHMM), not the schema ISO type.
Evidence
Real author checkpoint on CPU, populated specimen output and empty blank control. See src/probe/reports/2026-09-30-format-expansion/mothbot/report.json for exact invocation and scope. This is a format probe, not an accuracy evaluation. Runs the original _save_result after OBB inference; excludes thumbnail generation, blur enrichment and later pipeline stages.