Experiment trackers and trainers - Githubissues

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

Experiment trackers and trainers #156

Closed orperel closed 1 year ago

orperel commented 1 year ago

This MR aims to refactor trainers to support the new config system.

Trainer ctors no longer receive >10 args, instead they take the minimal requires args: a config, a pipeline, a dataset, etc.
Usage of extra_args was removed and replaced with the config classes
New Tracker modules (trackers, metric board and refined api for offline visualizer) - this removes a lot of logging code from the trainers

The Tracker is a one stop for everything related to logging and visualizations.