isir

All format examples · Machine-readable schema · Convert outputs to COCO

One image, one detection. The box is pixel center-x, center-y, width, height with a bottom-left origin.

Example

Illustrative values, not predictions from a model. Unless stated otherwise, the image is 200 pixels wide and 100 pixels high.

{
  "image": {
    "file_name": "frame.jpg",
    "height": 100,
    "id": "frame-1",
    "width": 200
  },
  "instances": [
    {
      "bbox": [30, 60, 40, 40],
      "category_id": 0,
      "confidence": 0.8,
      "id": 0
    }
  ],
  "ir_id": 1,
  "ir_name": "ISIR"
}
Richer example: oriented geometry and metadata

Illustrative optional fields; ordinary conversions do not require all of them. This record still describes one 200 × 100 image.

The box is centered at [100, 50], with width 40, height 20 and a counterclockwise rotation of π/2 radians. Polygon coordinates use the same bottom-left origin. The example source reports an area of 800 square pixels; converters do not infer area from the box.

Model and inference fields record provenance; context describes the capture location and grouping. All names, settings, timestamps and coordinates here are illustrative.

Category 0 refers to a caller-supplied vocabulary. Namespaced extra_information retains source-specific details; target formats may not preserve this metadata.

{
  "context": {
    "crs": 4326,
    "group_id": 1,
    "group_name": "Example survey",
    "latitude": 55.68,
    "longitude": 12.57,
    "tag_list": ["daylight"],
    "timestamp": "2026-09-30T12:00:00.000Z"
  },
  "extra_information": {
    "example_pipeline": {
      "source_frame": 42
    }
  },
  "image": {
    "file_name": "frame.jpg",
    "format": "JPEG",
    "height": 100,
    "id": "frame-1",
    "width": 200
  },
  "inference": {
    "config": {
      "confidence_threshold": 0.25
    },
    "timestamp": "2026-09-30T12:00:01.000Z"
  },
  "instances": [
    {
      "angle": 1.5707963267948966,
      "area": 800,
      "bbox": [100, 50, 40, 20],
      "category_id": 0,
      "confidence": 0.8,
      "extra_information": {
        "example_detector": {
          "track_id": 7
        }
      },
      "id": 0,
      "polygons": [
        [
          [110, 30],
          [110, 70],
          [90, 70],
          [90, 30]
        ]
      ]
    }
  ],
  "ir_id": 1,
  "ir_name": "ISIR",
  "model": {
    "name": "Example detector"
  }
}

Convert to a standard format

Use Converter.convert with source="isir" 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: One ISIR record describes detections on exactly one image. Multi-image adapters return a collection of these records; dataset-level metadata may be retained separately from the per-image records.

Bbox: We use xywh: center coordinates and width/height - not normalized. Coordinate system has origin (0,0) at bottom-left, and top-right is (width, height).