ReconS

End-to-end RGBD pipeline: depth → PyCuVSLAM → nvblox.

Demo :

Demo-Image-to-Mesh.mp4

References:

• nvblox: https://nvidia-isaac.github.io/nvblox/index.html
• Neural reconstruction stereo (NuRec): https://docs.nvidia.com/nurec/robotics/neural_reconstruction_stereo.html
• Depth generation (Depth Anything TRT): https://github.com/ika-rwth-aachen/ros2-depth-anything-v3-trt

Environment

• Use the pycuvslam conda env (has cuvslam). Install the rest into the same env:

  conda activate pycuvslam
  python -m pip install "torch==<cu12x build>" -f https://download.pytorch.org/whl/torch_stable.html
  python -m pip install <nvblox_torch_wheel>  # matching CUDA build
  python -m pip install rerun-sdk opencv-python numpy

  Replace <cu12x build> and wheel path with the CUDA build you have (cu12 wheels run on CUDA 13 drivers).
• NVIDIA driver: PyCuVSLAM requires a driver exposing CUDA ≥12.6 (R560+; CUDA 13 is OK). Check with nvidia-smi.
• On WSL add GUI libs if needed and keep LD_LIBRARY_PATH=$CONDA_PREFIX/lib:/usr/lib/wsl/lib:$LD_LIBRARY_PATH so cuvslam loads the WSL libcuda shim instead of older toolkit stubs.
• If conda is missing after install, run ~/miniconda3/bin/conda init bash, then source ~/.bashrc (or reopen a shell) so conda is on your PATH.

Data layout (per sample)

data/sample_xxx/
  iphone_mono/             # RGB frames (0000001.png ...)
  iphone_mono_depth/       # Depth frames aligned to RGB (uint16, mm)
  iphone_calibration.yaml  # Pinhole K
  timestamps.txt           # frame,timestamp_ns

Outputs land alongside the sample (or next to your provided RGB folder):

• cuvslam_poses.tum and cuvslam_poses_slam.tum (SLAM enabled by default)
• nvblox_out/mesh.ply (+ voxel exports)
• depth folder is regenerated if missing.

One-shot pipeline

python3 run_full_pipeline.py \
  --rgb-dir data/sample_20260119_i4/iphone_mono \
  --calibration data/sample_20260119_i4/iphone_calibration.yaml \
  --timestamps data/sample_20260119_i4/timestamps.txt \
  --nvblox-mode colormesh \
  --nvblox-ui   # drop if headless

This runs depth generation, PyCuVSLAM, builds nvblox artifacts, and runs nvblox with Rerun UI when --nvblox-ui is set. You can still pass --dataset <folder> to use the default filenames (iphone_mono, iphone_calibration.yaml, timestamps.txt), but --dataset is optional—any folder names work when you provide the three paths. SLAM runs by default: both odometry (pycuvslam_poses.tum) and SLAM (pycuvslam_poses_slam.tum, CameraTrajectory_slam.csv) are saved, and nvblox consumes the SLAM trajectory unless you pass --use-odom-poses (or disable SLAM with --disable-slam).

Individual runners

See pipelines/README.md for per-step commands (run_pycuvslam_rgbd.py, run_nvblox.py, stereo variant, etc.).