NVIDIAGameWorks / kaolin-wisp

NVIDIA Kaolin Wisp is a PyTorch library powered by NVIDIA Kaolin Core to work with neural fields (including NeRFs, NGLOD, instant-ngp and VQAD).
Other
1.45k stars 133 forks source link

Add Weights & Biases logging #75

Closed soumik12345 closed 1 year ago

soumik12345 commented 1 year ago

This PR adds support for experiment tracking using Weights & Biases. In order to track training and validation metrics, render 3D interactive plots, reproduce the configurations and results, and many more features in a Weights & Biases workspace just add the additional flag --wandb_project <your-project-name> when initializing the training script.

Complete list of features supported by Weights & Biases:

The full list of optional arguments related to logging on Weights & Biases includes:

We also wrote a detailed report on Variable Bitrate Neural Fields and its usage with kaolin-wisp. The report was published in our blog Fully-Connected.

wandb-github-badge-gradient

orperel commented 1 year ago

Hi @soumik12345 .

Thank you so much for this beautiful PR! The VQAD report was also a delight to read. I've tested your changes locally and some very minor modifications were required. We're excited to merge it, but before we do I should bring the following to your attention:

  1. We're about to release a version with some breaking changes to our main scripts, to make them more friendly for future applications (we'll make sure to merge your changes first and support them moving forward). Once we do that, we may list VQAD as a separate app with its own README. We would be more than happy to to add a link within the VQAD readme to your report and W&B project, if you're interested.

  2. We've been requested to ask external collaborators to sign a Collaborator License Agreement. That should protect your rights as a contributor, but we would need to have it signed before we merge the PR. I'll have the details added soon

soumik12345 commented 1 year ago

Hi @orperel Thanks a lot for your feedback on both the PR and the report.

  1. It would be our immense pleasure to include a link to the updated VQAD application in our report. We would edit the report to include an updated and in-depth tutorial train the model after the changes.

  2. As for the Collaborator License Agreement, I would sign it ASAP to prevent any further delay.

orperel commented 1 year ago

Hi @soumik12345 ! I've emailed you the CLA. I'm ready to merge as soon as you send it back. Thanks!

soumik12345 commented 1 year ago

Hi @soumik12345 ! I've emailed you the CLA. I'm ready to merge as soon as you send it back. Thanks!

Hi @orperel I have signed the CLA and sent you on the same email thread.