gmum / points2nerf

Points2NeRF
73 stars 9 forks source link

What is the hardware configuration best for this project? #7

Open shuyueW1991 opened 1 year ago

shuyueW1991 commented 1 year ago

Hi, I use RTX 3080 to train this project. It takes approximately 480 seconds to accomplish 1 epoch. Is this rate normal? what is a good gpu palette here? I noticed in the 'Limitation' of the paper that a chunk of time is required.

forestsen commented 1 year ago

Hi, I use RTX 3080 to train this project. It takes approximately 480 seconds to accomplish 1 epoch. Is this rate normal? what is a good gpu palette here? I noticed in the 'Limitation' of the paper that a chunk of time is required.

去年年底的时候花了半个月炼到 5000 个 epoch,效果依然很差。他这篇论文投 ICLR2023 没中。他论文的架构还需要优化。

Ideefixze commented 1 year ago

Hi. We trained our model on ShapeNet car category data set containing 2797 objects for ~2200 epochs on Tesla V100-SXM2 GPU, which in total required 300 hours of calculation. Each epoch took ~500 seconds, so your rate is okay if you use the same dataset.