All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
Addedfor new features.Changedfor changes in existing functionality.Deprecatedfor soon-to-be removed features.Removedfor now removed features.Fixedfor any bug fixes.Securityin case of vulnerabilities.
- A simple jupyter notebook (#55).
sdf_biasas an alternative way for SDF initialization inimplicit-volume(#57).- Automatically remove outliers with a small number of faces when extracting surfaces (#61).
- The implementation of ProlificDreamer (#74, #105).
- An experimental implementation of using Zero-1-to-3 for 3D generation from a single image (#71).
- Support mesh initialization for
implicit-sdf(#90). - Easy-to-use geometry conversion by
system.geometry_convert_from. This is used in the Magic3D and ProlificDreamer system and may inspire applications connecting multiple systems/algorithms (#105). - Support prompt debiasing and manual assignment of view-dependent prompts (#98).
- The implementation of Perp-Neg (#98).
- Support patch-based renderer (#154).
- Support 3D reconstruction from multi-view images and 3D editing based on InstructNeRF2NeRF/ControlNet (#119).
- Support NeuS/VolSDF volume renderer and the coarse stage of TextMesh (#162,#121).
- Gradio web interface (#183).
- Remove
trainerfrom the constructor arguments of prompt processors (#56). - Use a reparametrization trick for the SDS loss (#57).
- Make Magic3D coarse stage use analytic normal and orientation loss.
- Move the logic of getting text embeddings according to camera settings from prompt processors to guidance (#77).
- Remove
from_coarsefrom the Magic3D system. Usesystem.geometry_convert_frominstead (#105).
- Fix errors caused by empty rays (#152).
- Implementation of DreamFusion, Magic3D, SJC, Latent-NeRF and Sketch-Shape.
- Implementation of the geometry stage of Fantasia3D.
- Multi-GPU training support (#33).
- Mesh export, supporting obj with mtl and obj with vertex colors (#44).