Missed Detection Bias

Available

object_detection.missed_detection_bias

How evenly the model finds objects across classes, exposing classes that are disproportionately missed.

Contract

Field	Value
version	1.0.0
metric_type	pointwise
scorer_contract	per_row
direction	lower_is_better
entrypoint	aip_metrics_object_detection.answered.missed_detection_bias
target_kind	None

Required columns

• image_id
• label
• predictions

Accepted schemas

[
  {
    "name": "gdi_image_v1",
    "task_types": [
      "detection"
    ]
  }
]

Methodology

1. Deterministic - no LLM. Uses Resaro's cv-eval library (custom Jaccard IoU, default threshold 0.5).
2. A ground-truth box counts as found when a same-class prediction overlaps it above the IoU threshold.
3. Sum missed boxes per class and report total missed / total ground-truth, plus the per-class breakdown.

Score semantics

Scores capture the overall miss rate, with a per-class breakdown of what is missed. A low score means few objects are missed across classes; a high score means many are missed. Lower is better.

Worked example

4 ground-truth boxes with only 1 matched by a same-class prediction -> 3 missed -> miss rate = 3/4 = 0.75.

Configuration schema

{
  "properties": {
    "inverted": {
      "default": true,
      "title": "Inverted",
      "type": "boolean",
      "x-aip-param-role": "scoring_metadata"
    },
    "iou_threshold": {
      "default": 0.5,
      "title": "Iou Threshold",
      "type": "number"
    },
    "label_map": {
      "anyOf": [
        {
          "additionalProperties": {
            "type": "integer"
          },
          "type": "object"
        },
        {
          "type": "null"
        }
      ],
      "default": null,
      "title": "Label Map"
    },
    "max": {
      "default": 1.0,
      "title": "Max",
      "type": "number",
      "x-aip-param-role": "scoring_metadata"
    },
    "metric_name": {
      "default": "missed_detection_bias",
      "title": "Metric Name",
      "type": "string",
      "x-aip-param-role": "scoring_metadata"
    },
    "min": {
      "default": 0.0,
      "title": "Min",
      "type": "number",
      "x-aip-param-role": "scoring_metadata"
    },
    "weight": {
      "default": 1.0,
      "title": "Weight",
      "type": "number",
      "x-aip-param-role": "scoring_metadata"
    }
  },
  "title": "InitializationConfig",
  "type": "object"
}

Execution

{
  "emits_metric_family": false,
  "function_name": null,
  "max_concurrency": null,
  "processing_kind": "cpu",
  "stream_batch_size": null,
  "timeout_seconds": 60
}

Complete manifest

accepts:
- name: gdi_image_v1
  task_types:
  - detection
config_schema:
  properties:
    inverted:
      default: true
      title: Inverted
      type: boolean
      x-aip-param-role: scoring_metadata
    iou_threshold:
      default: 0.5
      title: Iou Threshold
      type: number
    label_map:
      anyOf:
      - additionalProperties:
          type: integer
        type: object
      - type: 'null'
      default: null
      title: Label Map
    max:
      default: 1.0
      title: Max
      type: number
      x-aip-param-role: scoring_metadata
    metric_name:
      default: missed_detection_bias
      title: Metric Name
      type: string
      x-aip-param-role: scoring_metadata
    min:
      default: 0.0
      title: Min
      type: number
      x-aip-param-role: scoring_metadata
    weight:
      default: 1.0
      title: Weight
      type: number
      x-aip-param-role: scoring_metadata
  title: InitializationConfig
  type: object
dependencies: []
description: How evenly does the model find objects across all classes?
direction: lower_is_better
display_name: Missed Detection Bias
entrypoint: aip_metrics_object_detection.answered.missed_detection_bias
execution:
  emits_metric_family: false
  function_name: null
  max_concurrency: null
  processing_kind: cpu
  stream_batch_size: null
  timeout_seconds: 60
kind: metric
manifest_version: '1'
max_prompt_slots: 0
metric_metadata:
  methodology:
  - Deterministic - no LLM. Uses Resaro's cv-eval library (custom Jaccard IoU, default
    threshold 0.5).
  - A ground-truth box counts as found when a same-class prediction overlaps it above
    the IoU threshold.
  - Sum missed boxes per class and report total missed / total ground-truth, plus
    the per-class breakdown.
  score_semantics: Scores capture the overall miss rate, with a per-class breakdown
    of what is missed. A low score means few objects are missed across classes; a
    high score means many are missed. Lower is better.
  summary: How evenly the model finds objects across classes, exposing classes that
    are disproportionately missed.
  worked_example: 4 ground-truth boxes with only 1 matched by a same-class prediction
    -> 3 missed -> miss rate = 3/4 = 0.75.
metric_type: pointwise
name: object_detection.missed_detection_bias
partition_types: []
required_columns:
- image_id
- label
- predictions
required_kinds: []
scorer_contract: per_row
target_kind: null
unsupported_trace_shapes: []
version: 1.0.0