train time

[mowangmodi]

Hello, how long does it take you to train a model (such as brandenburg_gate), and what hardware configuration do you use for training?

[rover-xingyu]

We optimize all implementations for 600, 000 iterations with a batch size of 1024 on 4 A100s, where Ha-NeRF and NeRF-W take 20 and 18 hours respectively.
These models spend the same inference time once the appearance vector is given.
Ha-NeRF takes a few milliseconds to obtain the appearance vector from the CNN-based module, while NeRF-W needs minutes to optimize the vector.
Both methods require around 20 GB memory for training and 5 GB memory for inference.