HumanRig - A dataset for automatic rigging of 3D humanoid characters launched by Ali Gaode.

What is HumanRig?

HumanRig is a 3D humanoid character automatic rigging research project developed by the Alibaba team. It addresses the limitation of existing rigging techniques due to the lack of high-quality datasets by providing a large-scale, high-quality dataset and an innovative automatic rigging framework, thus promoting the automation of 3D character animation production. The HumanRig dataset contains 11,434 T-pose meshes, following a unified skeletal topology, and featuring diverse head and body proportions, filling gaps in existing datasets in terms of scale, diversity, and skeletal consistency. The automatic rigging framework uses a priori guided skeleton estimator (PGSE) and a mesh-skeleton mutual attention network (MSMAN) to achieve coarse-to-fine 3D skeleton joint regression and skinning weight estimation, generating characters suitable for animation production with superior performance compared to existing methods.

HumanRig's main functions

• Provide large-scale, high-quality datasetsHumanRig is the first large-scale dataset specifically designed for the automatic rigging of 3D humanoid characters, containing 11,434 high-quality AI-generated humanoid meshes. All models are rendered in T-pose and strictly adhere to industry-standard skeletal topologies, making them directly applicable to mainstream animation engines. This dataset boasts significant advantages in scale, diversity, and skeletal consistency, covering a wide range of character types from real people to cartoon characters and even anthropomorphic animals.
• Prior-guided skeleton estimator (PGSE)Based on 2D prior information projected into 3D space, a rough skeleton is initialized, which greatly reduces the complexity of binding tasks.
• U-shaped Point Transformer as a Mesh EncoderIt eliminates the reliance on edge information of 3D meshes, thus enhancing the binding robustness on complex meshes.
• Mesh-Skeleton Mutual Attention Network (MSMAN)By fusing mesh and skeleton features in a high-level semantic space, joint optimization of skeleton construction and skinning was achieved.

HumanRig's technical principles

• HumanRig dataset construction:
  • Generate diverse 2D images: Generate diverse T-pose character images from text descriptions using AI-driven 2D image generation technology.
  • Generate high-quality 3D meshesUse tools such as InstantMesh and Unique3D to convert 2D images into high-quality 3D meshes.
  • Filtering and OptimizationFrom 17,268 initial meshes, 14,662 high-quality models were selected, and skeleton rigging was performed using the Mixamo semi-automatic tool, resulting in 11,434 high-quality rigged models.
• Automatic binding process:
  • Skeleton initializationGenerate a rough skeleton using the PGSE module.
  • Feature extractionSkeleton and mesh features were extracted using an MLP-based skeleton encoder and a U-shaped Point Transformer, respectively.
  • Feature fusion and optimizationBy fusing skeleton and mesh features using the MSMAN module, we can achieve coarse-to-fine skeleton joint regression and skin weight estimation.
  • Generate animated charactersFinally, the optimized skeleton and skin weights are combined to generate a character that can be used for animation production.

HumanRig's project address

• Project official website:https://c8241998.github.io/HumanRig/
• arXiv technical paper:https://arxiv.org/pdf/2412.02317

Application scenarios of HumanRig

• Game developmentIn game development, HumanRig's automatic rigging technology can significantly reduce the time and cost of character animation production, and performs exceptionally well when dealing with complex character models, such as characters with complex clothing or accessories.
• Film and television productionIn the film and television industry, HumanRig's automatic rigging technology can quickly generate high-quality character rigging, significantly improving production efficiency.
• Virtual Reality (VR) and Augmented Reality (AR)In virtual reality and augmented reality applications, real-time interactive character animation is key to enhancing immersion. HumanRig's automatic rigging technology provides real-time skeletal animation support for virtual characters, ensuring natural and fluid character movements.
• 3D Digital HumanThrough automatic binding technology, Amap can quickly generate personalized 3D digital humans, providing users with a more interactive and fun navigation experience.