231106

Highlights

• Added reflection handling in texture model (neuralangelo-colmap_sparse-wreflection.yaml)
• Optimized mesh extraction (extract_mesh.py, geometry.py)

Change Log

Configurations

• Updated neuralangelo-colmap_dense.yaml
  • Changed the value of img_downscale from 1 to 2 (commit: efa1853)
  • Changed the value of radius from 2.5 to 1.5 (commit: efa1853)
• Updated neuralangelo-colmap_sparse-wmask.yaml
  • Changed the value of val_check_interval from 5000 to 10000 (commit: 5f9b7f5)
  • Changed the value of limit_val_batches from 2 to 1 (commit: 5f9b7f5)
• Added neuralangelo-colmap_sparse-wreflection.yaml (commit: aac5d75)
• Updated neuralangelo-colmap_sparse.yaml
  • Changed the value of n_test_traj_steps from 30 to 6 (commit: f44fe2f)
  • Changed the value of val_check_interval from 5000 to 10000 (commit: f44fe2f)

Export Module

• Updated export.py
  • Added support for exporting meshes with diffuse texture (commit: a864078)

Models

• Updated geometry.py
  • Improved the performance and stability of the Marching Cube algorithm (commit: 972c6d5)
• Updated texture.py
  • Added support for reflection in the VolumeDualColorV2 model (commit: 1b356dc)

Utilities

• Updated network_utils.py
  • Updated the ProgressiveBandHashGrid module to set the initial level based on the configuration (commit: 54c66f8)
• Updated utils.py
  • Added the reflect function for reflecting view directions (commit: 4e03c41)

Miscellaneous

• Updated requirements.txt to use CuMCubes instead of PyMCubes (commit: fbeaa29)
• Updated mixins.py in the utils module to include vertex normals in the saved mesh (commit: a3955b6)

231030

Highlights

• Added support for spherical harmonics (SH) radiance field
• Added SH-neuralangelo variant using spherical harmonics
• Added hloc support for better SFM
• Added foreground sampling and masking options
• Added scripts for filtering sparse points with mask

Change Log

New Features

• Added spherical harmonics (SH) radiance field in texture model (f49dd6d)
• Added SH-neuralangelo variant using spherical harmonics (35eacf3)
• Added hloc support for better SFM (1d586d1, 05bc950)
• Added foreground sampling and masking options (5f9b7f5, 806a1c2)
• Added model for filtering sparse points with mask (fc21724)

Enhancements

• Increased network capacity for better quality (bf9f9b9, f44fe2f)
• Added progressive training for SH model (35eacf3)
• Added auxiliary SH prediction loss for regularization (300962e)

Bug Fixes

• Fixed oversmoothing issue on flat regions (e2eab0d)
• Improved robustness to noisy SFM points (e2e