I'm new to the 3D reconstruction task and I noticed that most other methods require 8 GPUs with 1-3 days for training. For example, NeuralRecon's inference time (90ms as in your paper) is pretty similar to SimpleRecon's (72ms), but their training requires much more epochs so takes a much longer time. Could you share any insights behind the design makes SimpleRecon so efficient and effective (higher reconstruction quality)?
Dear Authors:
Thank you so much for sharing this amazing work!
I'm new to the 3D reconstruction task and I noticed that most other methods require 8 GPUs with 1-3 days for training. For example, NeuralRecon's inference time (90ms as in your paper) is pretty similar to SimpleRecon's (72ms), but their training requires much more epochs so takes a much longer time. Could you share any insights behind the design makes SimpleRecon so efficient and effective (higher reconstruction quality)?
Any information is greatly appreciated!
Sincerely, Ziyue Feng