search

guanyingc / DeepHDRVideo

HDR Video Reconstruction: A Coarse-to-fine Network and A Real-world Benchmark Dataset (ICCV 2021)
https://guanyingc.github.io/DeepHDRVideo
Other
99 stars 19 forks source link

Question about Data augmentation #26

Open chaolistat opened 1 year ago

chaolistat commented 1 year ago

In the paper, you mentioned adding noise to the training data and perturbeding the tone. May I know where the code for this part is? I couldn't find them. Thank you

chaolistat commented 1 year ago

Also, I would like to ask what the purpose of the following code is. I don't quite understand, thank you very much! image

guanyingc commented 1 year ago

Hi,

  1. Add noise on reference frame: https://github.com/guanyingc/DeepHDRVideo/blob/b9686676122f1eb879673448c78f853ad8c523a3/models/hdr2E_model.py#L128 2.perturb the tone in: https://github.com/guanyingc/DeepHDRVideo/blob/b9686676122f1eb879673448c78f853ad8c523a3/models/hdr2E_model.py#L132

  2. The attached code generates a "new maximum value" for re-exposure. For example, if the value of new_anchor is 0.97, after the last two line of codes, values that larger than 0.97 will be cliped.

chaolistat commented 1 year ago

Hi,

  1. Add noise on reference frame: https://github.com/guanyingc/DeepHDRVideo/blob/b9686676122f1eb879673448c78f853ad8c523a3/models/hdr2E_model.py#L128

    2.perturb the tone in: https://github.com/guanyingc/DeepHDRVideo/blob/b9686676122f1eb879673448c78f853ad8c523a3/models/hdr2E_model.py#L132

  2. The attached code generates a "new maximum value" for re-exposure. For example, if the value of new_anchor is 0.97, after the last two line of codes, values that larger than 0.97 will be cliped.

Thanks! But for 2. I still don't quite understand why a "new maximum value" is done. And is the way the paper simulates over/ubder exposure consistent with real life?

chaolistat commented 1 year ago

Or could you tell me which part of the paper a "new maximum value" corresponds to? thanks