Result Object Structure

Understand the structure of prediction results returned by DeGirumJS.

The AIServerModel and CloudServerModel classes return a result object that contains the inference results from the predict and predict_batch functions.

Example:

let someResult = await someModel.predict(image);
console.log(someResult);

For example, the result can be structured like this:

{
    "result": [
        [
            { "category_id": 1, "label": "foo", "score": 0.2 },
            { "category_id": 0, "label": "bar", "score": 0.1 }
        ],
        "frame123"
    ],
    "imageFrame": imageBitmap
}

Accessing the Result Data

• Inference Results: Access the main results using someResult.result[0].

• Frame Info / Number: Get the frame information or frame number using someResult.result[1].

• Original Input Image: Access the original input image with someResult.imageFrame.

---

Inference Result Types

The inference results can be one of the following types:

• Detection: Contains bbox, category_id, label, score, optional landmarks and mask for instance segmentation and pose estimation.

• Classification: Contains label and score for single-label classification.

• Multi-Label Classification: Contains classifier name and a results array with multiple { label, score } entries.

• Segmentation Mask: Contains shape and data, where data is a flat array (or Uint8Array) representing class IDs per pixel.

• Pose Detection: Contains landmarks array with each landmark having category_id, landmark coordinates, optional score, label, and connectivity information.

---

Example Results

Detection Result

Detection results include bounding boxes (bbox) along with category IDs, labels, and confidence scores. Optionally, masks and landmarks may be included for segmentation and pose.

[
  {
    "category_id": 2,
    "label": "person",
    "score": 0.98,
    "bbox": [0.1, 0.2, 0.4, 0.8]
  },
  {
    "category_id": 3,
    "label": "dog",
    "score": 0.87,
    "bbox": [0.5, 0.3, 0.9, 0.7],
    "mask": {
      "width": 128,
      "height": 128,
      "data": "<RLE string>"
    }
  }
]

Classification Result

Single-label classification results contain a label, score, and category_id.

[
  {
    "category_id": 0,
    "label": "dog",
    "score": 0.76
  }
]

Multi-Label Classification Result

Multi-label classification results include a classifier name and an array of label-score pairs.

[
  {
    "classifier": "scene_tags",
    "results": [
      { "label": "beach", "score": 0.85 },
      { "label": "sunset", "score": 0.65 }
    ]
  }
]

Segmentation Mask Result

Semantic segmentation results provide a shape and a data array, which is typically returned as a Uint8Array.

[
  {
    "shape": [1, 256, 256],
    "data": [0, 1, 1, 0, /* ... repeated for 256*256 entries ... */]
  }
]

Pose Detection Result

Pose detection results include landmarks for each detected person, where each landmark has coordinates, optional score and label, and connectivity information.

[
  {
    "category_id": 0,
    "label": "person",
    "score": 0.98,
    "bbox": [0.1, 0.2, 0.4, 0.8],
    "landmarks": [
      {
        "category_id": 0,
        "landmark": [0.15, 0.25],
        "connect": [1],
        "label": "nose",
        "score": 0.99
      },
      {
        "category_id": 1,
        "landmark": [0.15, 0.35],
        "connect": [0],
        "label": "left_eye",
        "score": 0.97
      }
    ]
  }
]

---