bennyguo / instant-nsr-pl

Neural Surface reconstruction based on Instant-NGP. Efficient and customizable boilerplate for your research projects. Train NeuS in 10min!
MIT License
857 stars 84 forks source link

OOM 16384GB after I tried to add a Linear。。。。。。。。。 #103

Open StarsTesla opened 1 year ago

StarsTesla commented 1 year ago

Hi, there, I tried to modified the Volume Density Class: I separate the encoding and network, work with density as the first element of the feature, it works fine. But, when I tried to add a same network, now it comes to: a encoding, a network, and a density network(input feature, out one dimension density), the network only output feature, and then input to density network, then, OOM 16384GB, which amazing me. I guess it was the problem of occupancy grid, but, why? that's only one hidden layer, and 64 dimension in it. If I remove it ,just encoding and feature, that works totally fine.