FlexiSMPL

Abstract

FlexiSMPL is an interactive 3D visualization framework that provides real-time, measurement-based control of SMPL body shapes through an intuitive 3D interface. The system enables users to manipulate 23 anthropometric measurements (chest width, hip depth, etc.) via interactive sliders while observing immediate changes in a fully navigable 3D body model. Built on Open3D and PyTorch, FlexiSMPL maintains the semantic measurement-to-beta regression approach of LocalSMPL while introducing smooth 3D interaction, wireframe visualization modes, and optimized rendering performance. The system serves researchers, designers, and developers requiring precise anthropometric control with immediate visual feedback.

Introduction

The SMPL (Skinned Multi-Person Linear) model has become a standard for 3D human body representation in computer vision and graphics applications. While some other implementations provided measurement-based shape control through 2D front/side projections, many applications require full 3D visualization capabilities for comprehensive shape analysis and manipulation.

FlexiSMPL addresses this limitation with a complete 3D visualization pipeline. The system utilizes measurement-to-beta regression methodology along with an interactive 3D mesh viewing, real-time shape updates, and advanced visualization modes. Users can manipulate body measurements through an intuitive slider interface and immediately observe changes in a fully interactive 3D environment supporting rotation, zoom, and pan operations.

System Architecture

The system architecture emphasizes performance optimization through one-time mesh topology computation, vertex-only updates during shape changes, and continuous event polling for smooth 3D interaction. This approach ensures responsive interaction during shape transformations while maintaining the semantic meaning of anthropometric measurements.

  flowchart TD
    A["User Input
Measurement Sliders"] -->|" Measurement "| B["SMPL Controller"]
    B --> C["Measurement-to-Beta
Regression Model"]
    C -->|" Shape Parameters "| D["SMPL Model
PyTorch"]
    D -->|" 3D Vertices "| E["3D Viewer
Open3D"]
    E --> F["Interactive 3D Display
Mouse Controls"]
    G["Wireframe Toggle"] --> E
    H["Matplotlib Timer
Continuous Polling at 60FPS"] --> E
    E --> I["Rendering Optimizations"]
    I --> J["One-time Topology Setup
triangles computed once"]
    I --> K["Vertex-only Updates
no mesh recalculation"]
    I --> L["Immediate Visual Style
wireframe toggle"]
    style A fill: #e1f5fe
    style E fill: #f3e5f5
    style D fill: #fff3e0
    style C fill: #e8f5e8
    style I fill: #fff8e1

Key Components

Measurement Interface: Interactive sliders for 23 anthropometric measurements with real-time value conversion and immediate shape updates via callback functions.
SMPL Model: Uses pre-trained measurement-to-beta regression models for semantic shape control, generating new vertex positions through standard SMPL forward pass.
3D Visualization Engine: Open3D-based renderer implementing optimized mesh updates, native wireframe mode support, and coordinate frame visualization with customizable camera controls.
Continuous Interaction System: Runs at 60 FPS to ensure smooth 3D mouse interaction through continuous event polling and renderer updates.
Rendering Optimizations: One-time mesh topology computation (triangles set once), vertex-only updates during shape changes, and immediate visual style application for wireframe toggling.

Technical Specifications

SMPL Support: 300-PC shape space with male/female/neutral model variants
Performance: ~60 FPS interaction rate with sub-50ms measurement update latency
Measurements: 23 semantic body measurements (widths, depths, circumferences, lengths)
Visualization: Real-time 3D mesh rendering with wireframe toggle and full mouse navigation
Architecture: Modular design with separate controller, viewer, and UI components

BibTex