Inquiry about position.

[beautifulchoi]

Hello authors,

First of all, thanks for your open source contribution.

While reading the paper, I would like to know whether the model can also render well on the back view of the person if selecting the position just behind the person, just for my curiosity.

Also, how long does it take to train the final model using the same settings in the paper?

Thank you for your opinion in advance.

[KEH0T0]

Thank you for your interest!
Back view rendering: The model can attempt to generate from a position behind the person, but the quality may degrade significantly. Since our training data is predominantly front-facing egocentric views, the model has limited exposure to back-view scenarios and may not generalize well to that configuration.
Training time: With the same settings described in the paper, training takes approximately 24 Hours on 8 H200 gpus without validation and save checkpoints.
Hope this helps!