Why scenes so dark.

[chenkang455]

Hi @Pinocchio-Jia , thanks for your nice work and efforts for sharing this work. I discover that reconstructed scences are extremely dark, like TFP+NeRF and other comparative methods. Can you be kindly explaining it?

[Pinocchio-Jia]

Hi @Pinocchio-Jia , thanks for your nice work and efforts for sharing this work. I discover that reconstructed scences are extremely dark, like TFP+NeRF and other comparative methods. Can you be kindly explaining it?

Hello, because our training scenes are all low light intensity data, so the reconstruction results will look relatively dark, but compared with other methods, under the same light intensity, our visual effect will be better. https://drive.google.com/file/d/1tHPeRMnrEcN3l33B63bqezgASIIijDsT/view?usp=drive_link, you can use this link to obtain sample results figure, Among them, spikenerf-100k is the result obtained by training 100k times using the current version of the code, and the training weight of 100k times obtained is attached.

[chenkang455]

Hello, because our training scenes are all low light intensity data, so the reconstruction results will look relatively dark, but compared with other methods, under the same light intensity, our visual effect will be better.

Get it, thanks for your detailed response.

[chenkang455]

Hi @Pinocchio-Jia, I'm attempting to apply spike-nerf to my dataset using a raw binary spike stream, but it doesn't seem to be converging. In my implementation, I used short-tfp to replace the images in your dataset. Is this the correct approach? Additionally, could you please explain how you processed the raw spike stream to obtain the spike images included in your dataset? Thx a lot!

[chenkang455]

Should I adjust the lr or the snn noise parameters? Because i notice that your provided spike images have a large amount of noise, whose noise pattern might slightly differ from my data.