CharacterGen
Generate 3D character models from a single image
- Generate 3D character mesh from a single input image
- Calibrating poses using image conditional multi view diffusion model
- Generate detailed 3D models using a general sparse view reconstruction model based on transformers
- Using texture backprojection strategy to generate high-quality texture images
- Through quantitative and qualitative experimental evaluation, demonstrate its professional ability in generating high-quality 3D characters with shapes and textures
- Applicable to downstream binding and animation workflows
Product Details
CharacterGen is an efficient 3D character generation framework that can generate high-quality and consistent looking 3D pose unified character meshes from a single input image. It effectively calibrates input poses to a standardized form while preserving key attributes of the input image through a streamlined generation pipeline and image conditional multi view diffusion model, solving the challenges brought by diverse poses. It also adopts a general sparse view reconstruction model based on transformers, as well as a texture backprojection strategy to generate high-quality texture images.
