instant-ngp
instant-ngp copied to clipboard
When does it stop?
This is probably a simple mistake on my part, but I've been trying to run the nerf example on Google Colab, and it just keeps training. It seems to decrease in loss but is there a point where the training is supposed to stop? Do I need to do some sort of key command, command line argument, or do I simply just keep waiting? Thanks to anyone who can solve my stupidity. :)
Afaik it trains in realtime, it never stops. It does t give you an output once it's done, you see the output as it trains.
The problem is that the Google Colab example cannot use a gui because it is a command line interface, so how do I force it to eventually stop and output a mesh? Or am I stuck with skipping training entirely and immediately exporting the model?
you can try scripts/run.py
with the setting --n_steps
, the default value is 100000
Related to #427
@Vincentqyw It seems Step 10 (render video) of the colab, If I give n_steps > 2000 (which is the step for the first msgpack), it does not do anything while training. Loss seems to be the same. Did it train to the mentioned n_steps after loading the initial msgpack?