Training Hyperparams

[vidit98]

Thanks for your great work! Can you please specify the number of training steps in first and second stage training mentioned in the paper?

[eliphatfs]

50k per phase.

[vidit98]

Thanks! Could you also please tell the batchsize and number of objects in the dataset?

[eliphatfs]

256, and the size of Objaverse.

[vidit98]

Hi, I was wondering did you ever try training the model in single stage?

[eliphatfs]

The generalization was worse in an experimental, unreleased version of the model. I am not 100% sure if this is the same case for the v1.2 architecture.

[vidit98]

May I also know the SNR gamma you used for Min SNR strategy? Thanks again 🙏

[eliphatfs]

It is the default or paper value, perhaps 5.0