training time

[keaisg]

Thanks for sharing your wonderful work.
But I am curious about your training time for every part.

[Oliverbansk]

Thank you for your interest in our work!

I recommend training the DepthNet for 100 epochs, followed by synthetic training for all modules, also for 100 epochs, using the pretrained DepthNet weights. For the real-world self-supervised training, 80 epochs should be ideal.

Please note that the training times above are extended for optimal convergence, but you can adjust them based on your observations.

(The 700 epochs mentioned in the config are set to match the baseline RoboPose's configuration. However, this primarily represents the idea of "training until convergence.")