Computer vision · SDK

MediaPipe

Google

Integrate packaged perception tasks into device applications.

Official documentation Download development spec

GOOGLE MEDIAPIPE / DEVELOPER FIELD GUIDE

Choose an interface. Define a verifiable result.

Company or project context, ten technical entry points and a practical analysis of every reference.

2026-10-08 · Integration not tested

Company / project overview

Original summary of official company or project statements; use the source for the full original page.

Google MediaPipe · About
Project	MediaPipe is an open-source project within Google AI Edge.
Purpose	Provides reusable ML solutions for application developers.
Components	Tasks APIs and compatible models provide inference; Model Maker supports selected customizations and Studio supports evaluation.
Platform and lifecycle	Task and platform availability vary. Use current Solutions guides; legacy solution support differs.

Complete official About page ↗

About reference — TA analysis
Audience	Buyers, partners and product owners shortlisting an AI provider.
Our assessment	Use positioning and company history for initial fit. Obtain project-specific deployment, support and commercial terms separately.

Official project and developer sources reviewed; no provider API was called. Expected outputs below are our illustrative contracts, not measured results.

Top 10 technical entry points

Ten documented technical entry points, selected by task; not ten independent SDKs or an official ranking.

Choose by the work you need to complete
Interface	Purpose and audience
01 · Face DetectorAPI / SDK / developer guide	Locate faces and facial key points.Face region UI developers
02 · Face LandmarkerAPI / SDK / developer guide	Estimate facial landmarks and supported facial outputs.AR and animation developers
03 · Gesture RecognizerAPI / SDK / developer guide	Recognize supported hand gestures.Touchless interface developers
04 · Hand LandmarkerAPI / SDK / developer guide	Estimate hand landmarks and handedness.Hand interaction developers
05 · Holistic LandmarkerAPI / SDK / developer guide	Combine face, pose and hand landmark estimation.Whole-body interaction teams
06 · Image ClassifierAPI / SDK / developer guide	Assign categories using a compatible classification model.Image routing developers
07 · Image EmbedderAPI / SDK / developer guide	Represent images as numeric embeddings.Visual retrieval developers
08 · Image SegmenterAPI / SDK / developer guide	Separate image regions with a compatible segmentation model.Pixel-level UI developers
09 · Object DetectorAPI / SDK / developer guide	Locate and categorize model-supported objects.Object-based application teams
10 · Pose LandmarkerAPI / SDK / developer guide	Estimate body pose landmarks.Movement visualization teams

Reference analysis

Capabilities summarize official documentation. Constraints and proposed tests are our engineering assessment.

API / SDK / developer guide

Face Detector

Face Detector · TA
Target audience	Face region UI developers
Documented capability	Locate faces and facial key points.
Input → output	Image/frame → face boxes and key points.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Detection is not identity recognition.
Our suggested validation	Verify boxes against authorized annotated frames.

Official reference · Face Detector ↗

API / SDK / developer guide

Face Landmarker

Face Landmarker · TA
Target audience	AR and animation developers
Documented capability	Estimate facial landmarks and supported facial outputs.
Input → output	Image/frame → facial geometry; optional outputs depend on settings.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Geometry is not evidence of a person’s feelings or intent.
Our suggested validation	Check occlusion and coordinate mapping on target devices.

Official reference · Face Landmarker ↗

API / SDK / developer guide

Gesture Recognizer

Gesture Recognizer · TA
Target audience	Touchless interface developers
Documented capability	Recognize supported hand gestures.
Input → output	Image/frame → gesture categories and hand landmarks.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Only configured gesture vocabulary is meaningful.
Our suggested validation	Test unknown gestures without triggering a command.

Official reference · Gesture Recognizer ↗

API / SDK / developer guide

Hand Landmarker

Hand Landmarker · TA
Target audience	Hand interaction developers
Documented capability	Estimate hand landmarks and handedness.
Input → output	Image/frame → normalized and world landmarks, handedness.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Mirroring and coordinate systems must be explicit.
Our suggested validation	Compare mirrored and original input deliberately.

Official reference · Hand Landmarker ↗

API / SDK / developer guide

Holistic Landmarker

Holistic Landmarker · TA
Target audience	Whole-body interaction teams
Documented capability	Combine face, pose and hand landmark estimation.
Input → output	Image/frame → grouped body, face and hand landmarks.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Missing groups must remain missing rather than zero-filled people.
Our suggested validation	Verify partial visibility and group association.

Official reference · Holistic Landmarker ↗

API / SDK / developer guide

Image Classifier

Image Classifier · TA
Target audience	Image routing developers
Documented capability	Assign categories using a compatible classification model.
Input → output	Image → ranked categories and scores.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Category vocabulary is determined by the chosen model.
Our suggested validation	Include out-of-domain images and review thresholds.

Official reference · Image Classifier ↗

API / SDK / developer guide

Image Embedder

Image Embedder · TA
Target audience	Visual retrieval developers
Documented capability	Represent images as numeric embeddings.
Input → output	Image → feature vector.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Vectors are not semantic labels; compare compatible embeddings.
Our suggested validation	Evaluate relevant and irrelevant image pairs.

Official reference · Image Embedder ↗

API / SDK / developer guide

Image Segmenter

Image Segmenter · TA
Target audience	Pixel-level UI developers
Documented capability	Separate image regions with a compatible segmentation model.
Input → output	Image → category or confidence masks according to configuration.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Resized masks must be aligned before overlay.
Our suggested validation	Check pixel alignment and known-region overlap.

Official reference · Image Segmenter ↗

API / SDK / developer guide

Object Detector

Object Detector · TA
Target audience	Object-based application teams
Documented capability	Locate and categorize model-supported objects.
Input → output	Image/frame → boxes, categories and scores.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Pretrained classes do not automatically include factory defects.
Our suggested validation	Measure missed objects with domain-specific annotations.

Official reference · Object Detector ↗

API / SDK / developer guide

Pose Landmarker

Pose Landmarker · TA
Target audience	Movement visualization teams
Documented capability	Estimate body pose landmarks.
Input → output	Image/frame → normalized and world pose landmarks.
Execution environment	Selected Tasks SDK, compatible model and supported platform
Our decision constraint	Pose alone does not establish medical condition or safe operation.
Our suggested validation	Review occluded joints and frame-to-frame stability.

Official reference · Pose Landmarker ↗

Official documentation overview

Official documentation overview · Supporting reference analysis
TA	Developers and product owners selecting the supported technical path.
Our validation advice	Choose the operation, supported version and runtime before assigning implementation.

Official reference · Official documentation overview ↗

04 / INPUT · EXPECTED OUTPUT · ACCEPTANCE

Expected results before implementation

Illustrative object-detection contract for a selected Tasks model; other tasks return different structures. No camera or model was run.

Illustrative contract · no API call or model execution · not measured provider output

Object detected

Expected behavior — illustrative, not executed
Input	Authorized image and a compatible object model.
Expected result	ok; label, score and box in the declared coordinate system.
Acceptance	Model labels match the UI and boxes align after resizing.

No detection

Expected behavior — illustrative, not executed
Input	A valid image with no object above the chosen threshold.
Expected result	ok; detections: []; never reuse previous-frame boxes.
Acceptance	Empty state appears without an error banner.

Model mismatch

Expected behavior — illustrative, not executed
Input	A missing or incompatible task model.
Expected result	error; model initialization failed; data is null.
Acceptance	No fabricated predictions and a recoverable setup message.

Video sequence

Expected behavior — illustrative, not executed
Input	Frames with timestamps in the selected video/live mode.
Expected result	Results are associated with their input frame timestamps.
Acceptance	Reject or account for invalid ordering; do not promise a result for every live frame.

Map the expected result to your application

Review the scenarios below and map the actual SDK response into this internal contract. No credential, API call or live output is included. Replace example data only after your own integration test.

Expected output JSON — illustrative, not measured · MediaPipe-expected-results.json

{
  "schema_version": "1.0.0",
  "example": true,
  "provider": "Google MediaPipe",
  "status": "ok",
  "input_id": "sample-001",
  "data": {
    "task": "object_detector",
    "coordinate_system": "pixels-original-image",
    "detections": [
      {
        "label": "sample-object",
        "score": 0.82,
        "box_xywh": [
          12,
          18,
          80,
          60
        ]
      }
    ],
    "model_asset": "record-at-integration"
  },
  "error": null,
  "verification": "illustrative-only;not-provider-response"
}

Download specification, example and acceptance plan

Engineering handoff — our proposed contract and gates
Result states	ok means the selected operation returned a valid result; a valid empty result differs from error. Pending work remains pending until a terminal result. Do not infer real-world correctness from API success.
Reproducibility	Record input ID, library/model version, configuration, runtime and coordinate/score conventions. Examples contain invented sample values.
Acceptance — proposed	Use representative authorized samples plus empty, malformed and unavailable-runtime cases. Agree quality and latency targets before running tests. No measured accuracy, cost or speed is claimed.
Operation boundary	Illustrative object-detection contract for a selected Tasks model; other tasks return different structures. No camera or model was run.

Related route to assess: LiteRT

Development assessment

On-device perception benchmark

Development concept · integration not tested

Provider capabilities above are based on official documentation or repositories. The proposed product, inputs, deliverable and acceptance criteria below are our development assessment.

Integration pilot

Validate runtime, schemas or hardware together before estimating a deployable product.

Proposed inputs: A small authorized image dataset, task definition and reference annotations.
Proposed deliverable: A reviewable result with image references, labels or annotation state; keep original files.
Acceptance criterion: Compare the supported operation against a labeled sample; report errors and missing results separately.
Dependencies: Dataset access, image rights and a supported model or annotation project.

Development sequence

Confirm access to MediaPipe, license and the exact supported version.
Prepare the sample above and implement one documented operation for “On-device perception benchmark”.
Normalize the result with source, time and explicit error state; keep the provider response for review.
Run the acceptance criterion before estimating rollout effort or committing a customer deliverable.

How to validate demand

Record product views, documentation clicks, specification downloads and contextual hub clicks. These are event counts, not unique people or completed integrations.

This feasibility assessment uses implementation conditions. No traffic-based rank or delivery-time promise is assigned.

MediaPipe

Choose an interface. Define a verifiable result.

Company / project overview

Top 10 technical entry points

Reference analysis

Official documentation overview

Expected results before implementation

Object detected

No detection

Model mismatch

Video sequence

Map the expected result to your application

On-device perception benchmark

Development sequence

How to validate demand

Related products