How to render images with the same pose of GT from checkpoint.

[sunbeam-217]

Can anybody help me

I have trained my model , and I want to render some images to caculate PSNR, SSIM and LPIPS. The pose has been optimised in my metheod. I have tried "ns-render dataset " .Maybe there is a gap , The LIPIS is ok , but PSNR and SSIM is very bad .

[Parmisian]

I have the same problems in some of my datasets and the main issue usually turns out to be appearance/exposure related. For example,with scenes that have high exposure, the rendered result is quite bad. Check this out, ground truth vs nerfacto (cam optimization on/off did not change anything for me):

If you check your ns-eval results visually, maybe you can get an idea of what might be wrong but I dont have a solution for this yet

[sunbeam-217]

I have the same problems in some of my datasets and the main issue usually turns out to be appearance/exposure related. For example,with scenes that have high exposure, the rendered result is quite bad. Check this out, ground truth vs nerfacto (cam optimization on/off did not change anything for me):

If you check your ns-eval results visually, maybe you can get an idea of what might be wrong but I dont have a solution for this yet

Got it，Thank you very much