using with png

[vvubbalubba]

Hi! I'm trying to use this method with images i have only in PNG format. Is there any way to do this?

[Jamy-L]

Hi,

There is not really a straightforward way to make it work with any png image. Actually this algorithm is designed for raw images, so it would not work as well if the input has already been post-processed. 2 key elements are the noise distribution and the color space: denoising will not work so well if the noise distribution has been altered, and frame fusion must be done in the raw color space to work best.

Admitting that your png contains an image with minimal modification from the raw, there is also the question of channels. This algorithm was designed for mosaiced raw frames, so it can't work if the png is already and rgb image. You could either decimate the image to simulate mosaicing, or treat each channel as a separate image. We do have a grey_mode flag to process grey images, so you could process the R G and B channel by pretending that they are separate grey images.

I dont really know the nature of your png's, but this algorithm may not the best choice for your use case. If you still want to proceed, you need to run the code locally and modify the function that calls the pipeline. This means that you will need to manually normalise your image, get the noise profile and give the bayer pattern (in grey mode you can give whatever 2x2 matrice, it doesn't matter). If you have no idea of the noise profile, you can take alpha = 1.80710882e-4 and beta = 3.1937599182128e-6, and use the noise curve included in one folder of the project.

[vvubbalubba]

Thank you very much! <3С уважением, Фомина А. В.13.10.2023, 11:17, "Jamy Lafenetre" @.***>: Hi, There is not really a straightforward way to make it work with any png image. Actually this algorithm is designed for raw images, so it would not work as well if the input has already been post-processed. 2 key elements are the noise distribution and the color space: denoising will not work so well if the noise distribution has been altered, and frame fusion must be done in the raw color space to work best. Admitting that your png contains an image with minimal modification from the raw, there is also the question of channels. This algorithm was designed for mosaiced raw frames, so it can't work if the png is already and rgb image. You could either decimate the image to simulate mosaicing, or treat each channel as a separate image. We do have a grey_mode flag to process grey images, so you could process the R G and B channel by pretending that they are separate grey images. I dont really know the nature of your png's, but this algorithm may not the best choice for your use case. If you still want to proceed, you need to run the code locally and modify the function that calls the pipeline. This means that you will need to manually normalise your image, get the noise profile and give the bayer pattern (in grey mode you can give whatever 2x2 matrice, it doesn't matter). If you have no idea of the noise profile, you can take alpha = 1.80710882e-4 and beta = 3.1937599182128e-6, and use the noise curve included in one folder of the project.

—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you authored the thread.Message ID: @.***>