Clarification on Image Loading, Rendering for Evaluation, and Result Submission

[GuangyanCai]

Hi there,

Thanks for the great benchmark! We would love to try our method on it. I have a few questions though:

• I notice that there is a black border in every image. Is there a mask or a function to remove it? How was it treated during the previous evaluations?
• When generating the rendered images, are we supposed to uses the provided masks to replace the background with white color? I saw the example outputs, but I just want to confirm that it is true.
• If we are interested in reporting our results on the benchmark, what should we do?

Thanks again!

[zfkuang]

Hi Guangyan,

Thanks for the good questions! Here are (some of) the answers:

1. The black border is caused by our image processing step. Here, we need to shift the principle point of the camera to the center of image since both LLFF and Blender dataset are defined in this way. The black borders are always outside of our gt masks so no need to worry much about it. However, if you want to get rid of it without the mask, one way to do it is to center crop the image for a bit.
2. Any kinds of generated background should be fine - We only evaluate the area inside of the ground truth masks. But make sure to use our evaluation script for authentic evaluation scores.
3. We don't maintain a leaderboard at this stage, so feel free to report your score anywhere as long as it's calculated from our scripts.

Best, Zhengfei

[GuangyanCai]

Thank you Zhengfei, I don't have further questions.