Alpha

next week go through concept

As I mentioned yesterday, class next week will consist of Authoring Tool Design Doc presentations. Class on Mon (2/24) will be the Group 1 – 10 presentations while class on Wed (2/26) will be the Group 11-20 presentations.

Regarding the Design Doc presentations, here’s the format I would like you to follow for your 7 minute, max presentation. NOTE: you should have no more than 5 slides.

Slide 1 - Overview

• Name of your Authoring Tool

• Group Members

Slide 2 – Features and Capabilities

• Need for tool

• Bulleted list of main features and capabilities

  • The intent of this slide is to tell the audience what your tool does, what it takes as inputs, what it produces as output, etc.

Slide 3 – Technical Approach

• a VERY BRIEF description of the paper selected (i.e. Title, authors, SIGGRAPH year, main concepts)

Slide 4 - Implementation Strategy

• the intent of this slide is to describe your development approach

  • for example,

    • develop OpenGL test app to start, then develop C++ plugin

    • strictly MEL or Python scripting,

    • deep C++ Maya or Houdini integration

• Use of third party software

  • list specific libraries you might consider using
    • see CIS6600 Canvas site for list of libraries previously used in the past in CIS6600 projects.

---

Slide 3 – Technical Approach

• Paper Selected:
  • “RigNet: Neural Rigging for Articulated Characters” (Xu et al., SIGGRAPH 2020)
• Main Concepts
  • End-to-end graph neural network (GNN) pipeline for automated rigging
  • Predicts joint positions, skeleton connectivity, and skinning weights
  • Uses attention-driven clustering for joint detection and a minimum spanning tree to create a valid skeleton
  • Produces high-quality rigs across diverse shapes, from humanoids to fantasy creatures
• x
  • End-to-end graph neural network (GNN) pipeline for automated rigging
  • Predicts joint positions, skeleton connectivity, and skinning weights
  • Uses attention-driven clustering for joint detection and a minimum spanning tree to create a valid skeleton
  • Produces high-quality rigs across diverse shapes, from humanoids to fantasy creatures

---

Slide 4 – Implementation Strategy

• Overall Development Approach
  1. Write a Python-based plugin for Maya that handles data prep, UI, and rig generation
  2. Integrate pre-trained PyTorch (RigNet) models for inference (joint placement, connectivity, skin weights)
  3. Optional C++ components if certain geometry-processing steps need a speed boost
  4. Implement RigNet pipeline to train the model
  5. Integrate pre-trained PyTorch (RigNet) models for inference (joint placement, connectivity, skin weights)
  6. Develop a Python plugin Maya with PyQt that handles UI and rig generation
  7. Then develop the Python Plugin
• Use of Third-Party Software
  • PyTorch: for neural network inference
  • Torch Geometric: specialized GNN operations (EdgeConv, pooling, etc.)
  • Open3D: optional for volumetric geodesic or other geometry routines
  • Conda: environment management to ensure consistent dependencies
  • Maya Python API: to build a custom UI, convert meshes to GNN-friendly format, and create joints/skin clusters in Maya

Slide 4 - Implementation Strategy

• the intent of this slide is to describe your development approach

  • for example,

    • develop OpenGL test app to start, then develop C++ plugin

    • strictly MEL or Python scripting,

    • deep C++ Maya or Houdini integration

• Use of third party software

  • list specific libraries you might consider using
    • see CIS6600 Canvas site for list of libraries previously used in the past in CIS6600 projects.