Body Pix

Real-Time Human Body Part Segmentation

Identify and segment 24 distinct body parts in real time with sub-30ms latency. From virtual try-on to fitness analytics, Body Pix delivers pixel-perfect human understanding at production scale.

24 Body Parts
Segmentation Classes
<30ms
Latency
99.2%
Accuracy
Schedule Demo
Body Segmentation
Person Detected99.2%
24 Body Parts Labeled97.8%
Pose Estimation...
Real-Time Processing at 30+ FPS

Body Segmentation Use Case Simulator

See how real-time body segmentation supports virtual try-on, fitness coaching, and background replacement

Switch between deployment modes, inspect active body regions, and see how segmentation, pose, and overlay logic work together in production. The reference architecture shown here uses ML Kit, MediaPipe, and Cloud Run to support fast, responsive user experiences.

Apparel overlay

Retail Try-On

Torso

Segment the full body, isolate sleeves and torso, and keep garment overlays stable while the shopper rotates.

Active regionTorso · 99.1%
OutputComposite ready
99.1%

Primary garment alignment region

Garment drape stays anchored across shoulders, torso, and hips with clean edge refinement.

Selectable body regions

Live overlays

Edge feathering 8 pxOn
Occlusion recovery onOn
Virtual fabric physics liveOn
Session FPS 34On

How It Works

From Raw Video to Pixel-Level Understanding

Body Pix processes each video frame through a multi-stage pipeline that detects people, segments their body parts, estimates pose, and delivers structured output — all in under 30 milliseconds.

Person Segmentation

Instantly separate people from backgrounds with pixel-level precision. Works across diverse lighting conditions, complex backgrounds, and partial occlusions to deliver clean foreground masks for any downstream application.

99.2%
Pixel-level segmentation accuracy
  • Pixel-level foreground/background masks
  • Handles partial occlusions and overlaps
  • Robust across lighting conditions
  • Alpha matte generation for soft edges
  • Single-frame and temporal consistency modes

Body Part Labeling

Classify every pixel belonging to a person into one of 24 distinct body part regions. From head and torso to individual limbs and hands, get granular anatomical understanding for precise spatial reasoning.

24
Distinct body part classes
  • 24-class body part taxonomy
  • Head, torso, arms, hands, legs, feet regions
  • Left/right limb differentiation
  • Sub-pixel boundary refinement
  • Confidence scores per region

Pose Estimation

Detect and track 17 skeletal keypoints per person with high-fidelity joint localization. Combine pose data with segmentation masks for rich spatial understanding of human posture and movement.

97.5%
Keypoint detection accuracy
  • 17 skeletal keypoints per person
  • Joint angle and limb length estimation
  • Pose confidence scoring
  • Multi-person pose graphs
  • Temporal pose smoothing

Background Removal

Remove, replace, or blur backgrounds in real time without green screens. Produce broadcast-quality results suitable for video conferencing, content creation, and live streaming applications.

<15ms
Background swap latency
  • Real-time background replacement
  • Gaussian and bokeh blur modes
  • Virtual background insertion
  • Hair and edge refinement
  • Green-screen-free compositing

Multi-Person Tracking

Detect, segment, and independently track multiple people in the same frame. Maintain consistent identity assignment across frames even through occlusions and rapid movement.

15+
Simultaneous person tracking
  • Per-instance segmentation masks
  • Persistent identity assignment
  • Occlusion-aware re-identification
  • Crowd-density handling
  • Entry/exit detection

Real-Time Processing

Process video streams at 30+ FPS on standard hardware with optimized inference pipelines. GPU and CPU execution paths ensure deployment flexibility from edge devices to cloud infrastructure.

30+ FPS
Real-time processing speed
  • GPU and CPU inference paths
  • WebGL and WebAssembly support
  • Edge device optimization
  • Batch processing for offline workflows
  • Adaptive quality scaling

Complete Body Segmentation Suite

Enterprise-grade human body understanding powered by deep learning

Person Segmentation

Instantly separate people from backgrounds with pixel-level precision. Works across diverse lighting conditions, complex backgrounds, and partial occlusions to deliver clean foreground masks for any downstream application.

99.2%
Pixel-level segmentation accuracy
  • Pixel-level foreground/background masks
  • Handles partial occlusions and overlaps
  • Robust across lighting conditions
  • Alpha matte generation for soft edges
  • Single-frame and temporal consistency modes

Body Part Labeling

Classify every pixel belonging to a person into one of 24 distinct body part regions. From head and torso to individual limbs and hands, get granular anatomical understanding for precise spatial reasoning.

24
Distinct body part classes
  • 24-class body part taxonomy
  • Head, torso, arms, hands, legs, feet regions
  • Left/right limb differentiation
  • Sub-pixel boundary refinement
  • Confidence scores per region

Pose Estimation

Detect and track 17 skeletal keypoints per person with high-fidelity joint localization. Combine pose data with segmentation masks for rich spatial understanding of human posture and movement.

97.5%
Keypoint detection accuracy
  • 17 skeletal keypoints per person
  • Joint angle and limb length estimation
  • Pose confidence scoring
  • Multi-person pose graphs
  • Temporal pose smoothing

Background Removal

Remove, replace, or blur backgrounds in real time without green screens. Produce broadcast-quality results suitable for video conferencing, content creation, and live streaming applications.

<15ms
Background swap latency
  • Real-time background replacement
  • Gaussian and bokeh blur modes
  • Virtual background insertion
  • Hair and edge refinement
  • Green-screen-free compositing

Multi-Person Tracking

Detect, segment, and independently track multiple people in the same frame. Maintain consistent identity assignment across frames even through occlusions and rapid movement.

15+
Simultaneous person tracking
  • Per-instance segmentation masks
  • Persistent identity assignment
  • Occlusion-aware re-identification
  • Crowd-density handling
  • Entry/exit detection

Real-Time Processing

Process video streams at 30+ FPS on standard hardware with optimized inference pipelines. GPU and CPU execution paths ensure deployment flexibility from edge devices to cloud infrastructure.

30+ FPS
Real-time processing speed
  • GPU and CPU inference paths
  • WebGL and WebAssembly support
  • Edge device optimization
  • Batch processing for offline workflows
  • Adaptive quality scaling

Proven Impact on Visual Intelligence

Projected metrics for organizations using Body Pix segmentation

Segmentation Accuracy

Before
82%
After
99.2%
+17.2 pts

Processing Latency

Before
200ms
After
<30ms
85% faster

Background Removal Quality

Before
65%
After
96%
+31 pts

Multi-Person Detection

Before
3 people
After
15+ people
5x capacity

Edge Artifact Rate

Before
18%
After
1.2%
93% reduction

Integration Time

Before
6-8 weeks
After
3-5 days
90% faster

Example Scenarios

How Body Pix segmentation works in practice

Retail & E-Commerce

Virtual Try-On for Retail

Challenge

A major fashion retailer experienced 38% return rates on online apparel orders because customers could not visualize how garments would fit their body shape. Static size guides and 2D overlays failed to capture realistic draping and proportions.

Solution

Deployed Body Pix to segment individual body parts in real time via the customer's webcam. The 24-class body part map enables precise garment overlay that respects anatomical proportions, joint positions, and natural movement for a true-to-life virtual fitting room.

Results

52%
Reduction in return rate
3.2x
Increase in try-on engagement
28%
Higher conversion rate
$4.1M
Annual savings from fewer returns
"Customers finally trust what they see online. Our return rate dropped by half within the first quarter of deployment." — VP of Digital Commerce, Global Fashion Retailer
Health & Fitness

Fitness & Sports Analytics

Challenge

A fitness technology company needed to provide real-time form correction during home workouts. Existing pose-only solutions missed critical body positioning details like shoulder alignment and hip rotation that cause injuries.

Solution

Combined Body Pix body part segmentation with pose estimation to deliver pixel-level anatomical tracking. The system detects muscle group engagement, joint angles, and body alignment in real time, providing instant corrective feedback during exercise.

Results

67%
Reduction in form-related injuries
41%
Improvement in exercise effectiveness
<25ms
Feedback latency
4.8/5
User satisfaction rating
"Body Pix gave us the anatomical detail that pose estimation alone could never provide. Our injury rate dropped dramatically." — CTO, Connected Fitness Platform
Enterprise Communications

Video Conferencing Backgrounds

Challenge

A unified communications provider needed to offer virtual backgrounds that worked reliably across diverse hardware, lighting conditions, and user environments—without requiring green screens or dedicated GPUs on endpoint devices.

Solution

Integrated Body Pix person segmentation to deliver real-time background removal and replacement running entirely in the browser via WebGL. The solution handles hair detail, semi-transparent edges, and rapid movement without artifacts.

Results

30 FPS
Consistent frame rate on consumer hardware
94%
User satisfaction with background quality
0
Additional hardware required
12M+
Monthly active users
"We eliminated the green screen requirement entirely. Background replacement just works, on any laptop, in any room." — Director of Product, Enterprise Collaboration Suite

Implementation Timeline

From integration to production in 4 weeks

Week 1

Integration & Setup

  • SDK installation and API key provisioning
  • Camera input pipeline configuration
  • Basic person segmentation validation
  • Development environment optimization
Week 2

Feature Implementation

  • Body part labeling integration
  • Pose estimation pipeline setup
  • Multi-person tracking configuration
  • Background removal and replacement logic
Week 3

Testing & Optimization

  • Performance benchmarking across target hardware
  • Edge case testing (lighting, occlusion, crowds)
  • Latency optimization and quality tuning
  • User acceptance testing with pilot group
Week 4+

Production & Scale

  • Production deployment and monitoring setup
  • Auto-scaling configuration for traffic spikes
  • Analytics dashboard and accuracy tracking
  • Model fine-tuning based on production data

Frequently Asked Questions

Everything you need to know about Body Pix segmentation

Ready to See Bodies in a New Light?

Add real-time human body segmentation to your application with pixel-perfect accuracy and sub-30ms latency.