Kunhao-Liu / StyleRF

[CVPR 2023] StyleRF: Zero-shot 3D Style Transfer of Neural Radiance Fields
https://kunhao-liu.github.io/StyleRF/
143 stars 11 forks source link

Question about which 2 frames are selected to compute short/long range consistency #8

Closed Dhidbsixh closed 1 year ago

Dhidbsixh commented 1 year ago

Hi, very thanks for your work! In the paper, I would like to know how pairs of images are selected for calculating short/long range consistency in the paper, for example, in the LLFF dataset, are the images selected in the order in which they are named?

ruanchunqing7 commented 1 year ago

I have the same problem. Also, I have another problem. In the compute metrics, I don't see how to compute the RMSE, could you please tell me how to compute the RMSE?

Kunhao-Liu commented 1 year ago

Hi, we use two frames with close camera poses to compute the short-range consistency and frames with distant camera poses to compute the long-range consistency. The frame pairs are manually selected and differ in different scenes.

Kunhao-Liu commented 1 year ago

I have the same problem. Also, I have another problem. In the compute metrics, I don't see how to compute the RMSE, could you please tell me how to compute the RMSE?

Hi, you can refer to https://discuss.pytorch.org/t/rmse-loss-function/16540.

ruanchunqing7 commented 1 year ago

I have the same problem. Also, I have another problem. In the compute metrics, I don't see how to compute the RMSE, could you please tell me how to compute the RMSE?

Hi, you can refer to https://discuss.pytorch.org/t/rmse-loss-function/16540.

Thank you for your response. I mean, In your implementation code, how to calculate the RMSE in your paper, I do not see the relevant implementation.

image
ruanchunqing7 commented 1 year ago

Hi, we use two frames with close camera poses to compute the short-range consistency and frames with distant camera poses to compute the long-range consistency. The frame pairs are manually selected and differ in different scenes.

Thank you very much for your response. If we select deferent frames with you when compute the long-range consistency, will the metrics we get be very different from the LPIPS in the paper?

ruanchunqing7 commented 1 year ago

Hi, very thanks for your work! In the paper, I would like to know how pairs of images are selected for calculating short/long range consistency in the paper, for example, in the LLFF dataset, are the images selected in the order in which they are named?

Hello,

I have not found the implementation code corresponding to table 1, may I ask if you have found it? Thank you very much!