The inconsistency between the paper description and the checkpoint

[wangm-buaa]

In your paper, you mentioned the validation setup: "During training and evaluation, we adopt the original image splitting in Mega-NeRF and downsample the images by 6+ times from the original resolution." However, the rendering results of the published checkpoint seem to be downsampled by 4 times. What accounts for this discrepancy, and which should be considered the standard?

[AIBluefisher]

Hi, there is a typo in our paper. We follow all previous methods which downsamples images by 4 times in both training and evaluation.