Abdelrahman Abdelhamed

To train the model in the paper, you can remove `--cam` and `--iso` as these are used only to specify a single camera or ISO level, respectively.

> When using the released model for reproduce Marginal KL divergence between the generated and the real noise samples, i can only get 0.026( larger than 0.008 in paper). >...

Hi, This is just a convention. You may simply flip the figure and call the training direction "inverse" and the sampling direction "forward" and nothing would change, that is, the...

Just to clarify, in the noise flow paper, Figure 3, and in the code: - The training is in the inverse direction: noise distribution --> normal distribution. - The sampling...

> What's more, is there the pytorch implementation of Noise Flow? Not currently; I hope we can do it in the future.

Hi, There was a bug in the sampling code, it was not using the trained batch norm layers, this is fixed now. Also, I have added code for rendering the...

Please check out this potential fix: https://github.com/BorealisAI/noise_flow/commit/f9e2d29f350f8d59fd6be9af58ab41eeaafeb1db

The difference in image shape could be due to the sensor cropping area. This feature is not currently in the python version of the pipeline. Here is the equivalent in...

It is not necessary to get exactly the same sRGB images. This issue happens because different software or ISPs may render the images differently. For example, a different tone map...

Hi, You may need to install colour_demosaicing package. Try `pip install colour-demosaicing` or install it from https://pypi.org/project/colour-demosaicing/ I have updated the `requirements.txt` as well. Hope this helps!