Skip to content

Conversation

@MasahiroOgawa
Copy link

@MasahiroOgawa MasahiroOgawa commented Dec 2, 2025

Summary

  • Add uv virtual environment support as primary setup method
  • Update SIBR_viewers submodule for modern Linux compatibility (embree4, FFmpeg 5+, CUDA 12)
  • Add Ubuntu 24.04 support
  • Add CUDA submodules installation instructions
  • Improve network viewer reliability and documentation
  • Fix network_gui.py for better TCP handling and UTF-8 path support

Changes

uv Support

  • Add pyproject.toml with PyTorch CUDA 12.1 wheel configuration
  • Document uv as the recommended setup method

SIBR Viewers

  • Fix build for embree4, FFmpeg 5+, and CUDA 12
  • Add clear instructions for Linux SIBR viewer setup
  • Document that -s <dataset path> is required for network viewer

Network GUI Fixes

  • Fix TCP fragmentation handling in socket reads
  • Add SO_REUSEADDR to prevent "Address already in use" errors
  • Use UTF-8 encoding for non-ASCII dataset paths

Test plan

  • Tested uv sync and training on Ubuntu 24.04
  • Tested SIBR viewer build on Ubuntu 24.04
  • Tested network viewer live training visualization
  • Verified non-ASCII (Japanese) paths work with network viewer

🤖 Generated with Claude Code

Test result

Now it successfully displays Gaussians using SIBR remote viewer during training like this;
image

MasahiroOgawa and others added 7 commits November 26, 2025 17:54
Update submodule to include fixes for:
- embree4 support (with embree3 fallback)
- FFmpeg 5+ API compatibility
- CUDA 12 compilation fixes
- VideoUtils.hpp bug fix

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
- Add Ubuntu 24.04 to tested platforms
- Add uv sync as recommended setup method
- Keep Conda as alternative option
- Update SIBR viewer build to use Ninja by default
- Add note about avoiding Anaconda in PATH on Ubuntu 24.04

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
- Add instructions for running viewer after building from source
- Add PRIME offload workaround for hybrid GPU laptops (NVIDIA Optimus)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
Document the required step of installing diff-gaussian-rasterization,
simple-knn, and fused-ssim submodules after environment setup, since
these CUDA extensions require compilation and cannot be installed via
uv sync or conda alone.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
Clarify that after building from source, the SIBR install directory
is SIBR_viewers/install with executables in SIBR_viewers/install/bin/.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
- Fix network_gui.py socket read to handle TCP fragmentation properly
- Add SO_REUSEADDR to prevent "Address already in use" errors
- Use UTF-8 encoding for dataset paths (supports non-ASCII characters)
- Update README with clear instructions for live training visualization
- Document that -s dataset path is required for the network viewer

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant