Skip to content

Stage 5: Gaussian Splat Training #5

@CloudStrife7

Description

@CloudStrife7

What this stage does

Trains a 3D Gaussian Splat from your COLMAP data using LichtFeld Studio.
Expect 10–30 minutes on a modern NVIDIA GPU.

Tasks

  • 1. Confirm NVIDIA driver is 570 or newer: run nvidia-smi and check Driver Version
  • 2. Launch LichtFeld Studio binary from your download folder
  • 3. GUI: File > Import Dataset > select my_room_colmap/ folder
  • 4. OR CLI headless: ./LichtFeld-Studio -d ./captures/my_room_colmap -o ./output/my_room --strategy mcmc --max-cap 500000 --headless --bilateral-grid -i 30000
  • 5. Monitor loss curve in viewer — should decrease steadily (flat = bad capture data, see troubleshooting.md)
  • 6. Wait for training to complete at iteration 30000
  • 7. Locate output: ./output/my_room/point_cloud/iteration_30000/point_cloud.ply
  • 8. Open .ply in LichtFeld Studio viewer and visually confirm room is recognizable

Next step: When all 8 tasks are checked, close this issue and open Issue #6.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions