Anpei Chen

GPU with 8GB capacity may too small, previous I used 12GB 2080Ti, a larger GPU memory should be able to resolve this issue.

Hi XinWu, 1) currently the code is not supported "spheric_poses=False" since the near-far boundary must align with the cost volume construction, the cost volume is built on the real-world coordinate,...

I think there are splits [here](https://drive.google.com/file/d/1eDjh-_bxKKnEuz5h-HXS7EDJn59clx6V/view) or just to download the data example https://1drv.ms/u/s!AjyDwSVHuwr8zhAAXh7x5We9czKj?e=oStQ48

I remember that normalization doesn't matter a lot in practice.

The depth is not necessary for training, our depth is just for quantitative evaluation. Actually, you can convert your data into the DTU format if you are familiar with it....

Hi, I have checked that Ep(8) seems doesn't contain "dist" also. The trick of removing the marching size comes from the IBRNet for better generalizability, but it may not be...

> Sorry for the late reply. > > According to my understanding, delta in the second formula of Ep(8) is 'dist'. Am I getting this wrong? maybe you are using...

I used the official repo and here is my data loader: ``` import os import numpy as np from PIL import Image def load_dtu_data(basedir, downsample=1.0): def read_cam_file(filename, scale_factor, downsample): with...

Hi, yeah you definitely can use another resolution but must be 32*n since there are downsampling operations in the feature extraction

you may need to check the setting consistency between training and rendering, like the size of volume, padding