Skip to content

Integrate with Nerfstudio

Nerfstudio provides a API that allows for a simplified end-to-end process of creating, training, and testing NeRFs.

Open In Colab

If you can't wait, check out the results of this example Nerfstudio project for a preview of what's to come.

Comet SDKMinimum SDK versionMinimum Nerfstudio version
Python-SDK3.34.01.0.1

Start Logging

To log your Nerf training runs to Comet, use comet in the vis argument to your command line:

ns-train nerfacto --data data/nerfstudio/poster --vis comet --load-dir {outputs/.../nerfstudio_models}

Log automatically

By default, Comet will log the following items:

  • Metrics:
    • loss and other camera metrics
  • Parameters:
    • hyper-parameters defined for the training run
  • Images:
    • evaluation images throughout the trainig run

Visualize Progress in the Graphics Tab

Head over to the graphics tab in the Comet UI to see how your images are progressing during training.

nerf_animation

Log Video Files to Comet

Nerfstudio outputs video files after a training run is completed. Log them to an Existing Comet Experiment using

ExistingExperiment.log_video(output_video_path)

Then, head over to the Video Panel found in Comet's Built-In Panels section.

nerf_video

Apr. 25, 2024