bad performance on own data

[jinyummiao]

Hi, i have tried to training your model on own dataset (59 images in total, outdoor scenarios).
The trained depth priors seem proper:


However, the NeRF cannot be trained at all. Its loss is not decreased and the PSNR is nearly unchanged. And finally, the trained NeRF just renders totally black images for both depth and rgb:

The experimental setups are the same as the ones in the default configuration.
Could you give some advices on this problem? Thanks at advance :)